Background

I had toyed with the idea of going sober ever since the middle of college. Everyone has different relationships with substances, and mine with alcohol always felt like it should be temporary. As I got older and graduated, it felt like time to move on, and that anything more than a nice buzz should stay back in college. But month after month, I kept having nights where I got more drunk than intended, and I hated it. I could always control whether I drank — I had plenty of totally sober stretches in college — but when I drank, I was never the guy who could have one or two and stop (The em dashes were me, not ai). That’s what I hope would have changed by now. I also just hated how I acted when I was drunk.

My path to sobriety didn’t really take off until I got some concerning blood tests back in early 2025. The tests showed significant elevation in my ALT and AST, key liver enzymes used to detect liver stress and damage. My doctor mentioned that I should probably slow down on drinking if that’s what I had been up to, but that he didn’t think I had any kind of liver disease. But after seeing the results, and being more embarrassed than anything, I realized it was obvious what needed to change. I started to take my health much more seriously. I started to supplement with various things to help out my liver, eventually creating my own supplement stack for it.

These things did help, and they pushed me to really want to use moderation and to drink less. But weekend after weekend, I just couldn’t have one drink and call it a night. I wouldn’t even blame peer pressure, it was just that I never set a good boundary for myself. So, one morning that summer, I woke up and decided to cut it off entirely. Time to go sober.

Rules of my sobriety

Sobriety felt pretty scary at first. For the first few days I was terrified I’d give in, constantly thinking, “What if I drink again? What if I mess up?” I couldn’t even picture a long-term goal, only that I had to make it for two months.

But as the weeks went by, I started to get less scared. With each day in those first two months, I started to collect evidence that I could live my life and not drink. It started to feel more good than scary.

I started to settle in to things, thinking about them more practically. At first I had no idea what was in the bounds of sobriety versus out, so I banned alcohol of any kind. Mouthwash, even. Sobriety is less simple than it sounds, and there are a lot of technicalities. What about a “non-alcoholic” beer that actually has 0.1% or 0.5% alcohol? I would have to drink ~40 of those to equal one beer, but it’s still alcohol, right?

At first I was really strict about all this, but after a while, I realized that a single drop of alcohol in the NA beer was never the thing I was trying to escape. I was just trying to stop getting drunk. So I made new rules:

1. If it’s labeled “NA” (0.5% alcohol or less), I could drink it.
2. If something was classified as alcoholic, so >0.5%, it counted as breaking sobriety if I swallowed it.

That meant the trace alcohol from an NA beer was fine, and any kind of mouthwash or other alcohol was also fine if I rinsed out afterward. You might say these are permissive rules, and that’s fair, they worked for me. I was never in the territory of life-or-death alcoholism, so I didn’t feel like my sobriety needed to be that strict. I stuck to these rules for the entire year, so it was really important that they at least be manageable ones (Also, it feels both weird and pretty cool to say that I did that).

Another point of my sobriety was that it actually became threefold. I wanted it to feel like a real personal reset, so I brought in smoking (weed) and vaping too, not just drinking. Plenty of people don’t do these things to begin with, so I don’t think I’m “special” in any way, I just wanted to make it a little more challenging and remove the chance I’d replace one substance with another.

Just go longer

While my sobriety was threefold, it was only alcohol that made me afraid of relapse. Like I said, I told myself I was sober for now, not forever. I told everyone that I was doing it for two months, just to see where things went. I didn’t commit to going longer until I knew I could.

Cravings are like that. At first, they seem impossible to get rid of, but they fall off fast and don’t really come back. As I neared the halfway point of those two months, the daily panic about when I would drink next faded away, replaced by thoughts that felt more normal. It was around that time that I caught up with an old friend, who gave me what turned out being the single most important piece of advice I’d heard all year.

We got into chatting, and I explained my sobriety to him as well as the two-month goal. He had been sober for a pretty long time, so I imagine this didn’t sound like anything crazy. He asked why I settled on two months.

I don’t know, I told him. I just assumed that I’d be giving in at two months and wouldn’t be able to keep this going. I didn’t really want it to end in two months, I told him, but that I assumed it would.

“Why don’t you just go longer?”

I stared at him for a second. It never even occurred to me that I could just plan for longer.

But yeah, I could. Just go longer. So that’s exactly what I did. At the end of the two months I decided to keep going. Two months turned into six, and then at six months I said twelve. Now here we are, twelve months later. It’s crazy.

Changes I noticed

The changes in my body showed up fast, within 3 months. I noticed increased energy during the day and whiter eyes, which was a surprisingly great combo. I also worked on my general health and refined that supplement stack I mentioned. Around the same time as this, my cravings evaporated. By month three I no longer had any big urge to drink. I’d settled into the reality that I would just go to the bar with friends, sit my ass down, and order a soda or an NA beer. The lack of alcohol didn’t really manifest as a craving anymore, more like it was just a routine I had to follow. Get the NA option, nothing else.

Between the six-month to one-year period, the changes were more gradual, but these biomarkers trended to the best levels I’ve ever tested for:

• ALT
• AST
• HDL Cholesterol
• LDL Cholesterol
• Fasted Glucose
• Hemoglobin A1c

This was a really big accomplishment for me, since some of these were pretty high to begin with. I’m happy to say that, as of 2026, I have never had a better-functioning liver in my life, and that’s saying a lot since I have been getting these tests since I was seventeen!

What I still miss

I don’t miss being drunk. I actually saw that coming before I quit, as I could feel the hangovers kicking in before I would even get back home from a night out. Heavy drinking had already lost its appeal, which is why I had been slowly cutting the number of nights I drank per month. I don’t really miss that.

But something I do miss is that nice buzz you get after one, maybe two drinks. The state where you have a slightly elevated mood, and things seem a little more fun and playful. And that’s still where my head is today. The moments of drinking that I miss are pretty specific and infrequent. You’re at someone’s house, out on a date, at a birthday dinner, and one drink would be perfect to soften the mood. Without it, social stuff takes on a slightly different feeling. Not worse, but just a little different. Not drinking is fine, I’ll live, but I would like to experience alcohol again in moderation.

Drinking is also just part of culture here in the Midwest. We don’t have as much of the great outdoors as other regions, and we get some pretty bad winters, so a lot of times this is how we bond. Going sober made it pretty obvious to me that alcohol is ingrained into how people connect out here.

Would I recommend you do this?

People have asked me this, and I would recommend it! But I think you should do what’s right for you. Sobriety is awesome, but there’s something to be said for having the ability to use alcohol on occasion, to treat a drink the way you might have a sweet treat to celebrate a milestone. Or even to treat it like caffeine: you should be able to have a three-coffee day once in a while if that’s what you want. For people who can participate in this kind of thing responsibly, I would say to take advantage where you can.

In an ideal world, that’s how it would work. You’d use what you want in moderation, feel a specific way, and stop if you notice negative effects. But the last part takes a lot of introspection and honest self-assessment, and I don’t think it always plays out that way in real life. So for that reason, if you have issues with moderating your drinking, and you want to do something, there are tons of things you can try, from therapies to supplements and medications. And if you want resources for different things you can try, or different things that I’ve tried, I’m here for you. And if you want something more drastic, and you want to go sober full-stop, then yes, go for it! But if you don’t have a problem and you’re just asking whether you should go sober, I’d say you may get more satisfaction out of holding yourself to one or two drinks a night. Just start trying things and see what works!

Will I stay sober past one year?

I’m not sure yet. For me, it was about getting healthier and getting rid of my craving to drink. A year later, I’ve achieved a lot of what i was hoping for. I want to keep this same relationship with alcohol going forward, but with the ability to have a drink if I want one.

It’s also been incredible to see how much I got done with all the weekends I wasn’t going out. I refined my self-driving car project. I honed my health and got the best blood test results of my entire life. I worked on a passion project that graded two dozen phone chargers, continuing a Google engineer’s work. I created a half-dozen AI-assisted vibe coding projects, replacing stuff like my Spotify and my workout app with custom-coded alternatives that are 100% free to operate (article coming soon). A lot of stuff.

I plan to return to drinking casually one day. Maybe not just yet, but one day. And if I still can’t do the 1-2 drinks a night thing, I’ve got a year’s worth of sobriety showing me I can always do it again.

The baseline

I mentioned at the beginning that this all started because of some bad blood test results. But which ones specifically? The main markers came from a Comprehensive Metabolic Panel, something that I used to call this a liver panel, but realized isn’t quite right. A CMP covers more than the liver, so I filter it down to the values I care about:

• Total protein
• Albumin
• Globulin
• Albumin/globulin ratio
• Total bilirubin
• Alkaline phosphatase
• AST
• ALT

ALT and AST are the big ones for me because they are the main liver stress enzymes [0]. They don’t tell the whole story of liver function, but if you have been drinking and those numbers are high, they get your attention pretty quickly.

Albumin, total protein, globulin, and A/G ratio tell more of the surrounding story, helping to explain things like immune activity and whether the body is keeping its basic chemistry in a good place. (I sound like I know what I’m talking about, but most of this is from Google). Bilirubin and alkaline phosphatase are worth watching too, though bilirubin can bounce around for reasons that do not automatically worry me, and alkaline phosphatase is not purely a liver marker.

So while ALT and AST are my focus, the whole picture has to look good.

Now even inside of a normal range, the individual value is still important. After some research, I came across these generalized ‘trends’ for what makes something more optimal, even if it’s the normal range. Holding all else equal, if we are within the normal range, here’s what you want for each marker in a liver panel:

• Albumin: Higher is better
• A/G Ratio: Higher is better
• Total Protein: Higher is better
• Globulin: Dead middle is better
• ALT: Lower is better
• AST: Lower is better
• Alkaline Phosphatase: Lower is better
• Bilirubin: Makes no difference

This provides a way to differentiate test results that are close but still considered ‘within the normal range’. So if I’m comparing two test results that are within the normal range, clinically they may both be considered the same, but biologically one of them might be just slightly better.

One really cool piece of context to this is that I’ve been getting blood tests since I was seventeen. That is useful because I have my own younger-me data, which is more useful than a generic ‘normal range’ for these test results. I can look at what my numbers were before I spent a few years making worse decisions.

Now, here was seventeen-year-old me when he got his CMP:

Pretty good, honestly.

Then there was twenty-four-year-old me:

Yikes lol.

While my liver was still functioning generally well at the time of this test, the enzymes were extremely worrisome. As I mentioned before, this was taken after I graduated college. I spent nearly the entire calendar year at a bar, from Summer 2023 to Fall 2024. It was interesting since I had drank little in 2025 (at the time this test was done), but I stopped because the hangovers were getting too bad. So I knew any bad results this many months later were a sign of latent liver injury. And keep in mind, I was still drinking, just less frequently due to those hangovers.

Needless to say, after getting this test, I was motivated to finally take charge of my liver health.

Step one: research

The glutathione rabbit hole

I started to research more about glutathione, a supplement that could potentially help regenerate my liver. Funny enough, I had actually dealt with it before. Back in 2022, I went to a men’s clinic after my friend convinced me to get my blood tested like him. To optimize as many biomarkers as possible. Their main finding was that my liver was stressed with slightly elevated enzymes, and that I could inject glutathione if I wanted to. They put me on 400mg weekly.

The injections were really painful. It has to go straight into muscle, and it’s slightly acidic, so it stings while it goes in. I don’t have any hard data on whether those few injections did anything, but my liver was mostly normal at the time anyway. I ended up stopping it and forgetting about glutathione.

When I came back to it in 2025, I had one basic question: can I take this orally and avoid stabbing myself?

As I learned, plain glutathione has a bioavailability problem. One newer paper on glutathione analogues says native oral glutathione has bioavailability of <1% because it gets broken down enzymatically and does not absorb well through the gut [1]. That same paper tested modified versions of glutathione and found that one ‘N-methylated’ analog had much better half-life and oral bioavailability than native glutathione. Sounds promising, but I have no idea where someone would go to buy that.

The more practical thing I found was S-acetyl L-glutathione, usually shortened to SAG. It’s basically a modified glutathione molecule, forced onto an acetyl group that supposedly makes it more stable and helps it get into cells. This was probably the best oral composition out there.

The best human data I found was a small 2018 crossover study in eighteen healthy volunteers. It compared a single large oral dose of S-acetyl glutathione against regular glutathione. The study found higher plasma GSH Cmax and AUC after SAG than after the regular glutathione product, although some of the other results were messier and actually favored regular glutathione on some measures. Definitely not a slam dunk, but it was enough to make SAG look more plausible [2].

There was also a 2025 safety assessment in Food and Chemical Toxicology that looked at SAG for use in foods and dietary supplements. The summary I found reported no genotoxic signal, an acute oral LD50 above 2000 mg/kg in rats, and a 13-week rat NOAEL of 1500 mg/kg/day. That does not prove anything magical about how it absorbs, but it did make the safety profile look decent [3].

At that point, my conclusion was pretty simple: oral glutathione itself is probably not the move, but S-acetyl glutathione is at least interesting enough to try.

Step two: stop drinking

While the supplements helped, blood tests from the month afterward showed only moderate improvement. My liver enzymes were still elevated. The smart move was easy to understand but hard to do. I needed to stop drinking.

It was about three months of experimenting with supplementation before I fully accepted that I was not going to fix my liver while giving it the same thing that caused the problem. I had already been thinking about going sober for a while, so the bloodwork made that decision feel easier to make. It went from “maybe this would be a good reset” to “I don’t think my liver will go back to normal if I don’t do this”.

I made the decision to stop drinking. It was about 3.5 months after that first concerning blood test. My liver continued to improve in the months afterward, but the enzymes were still nothing incredible.

The stack

After three months sober, and 6.5 months into my journey of committing to improve my liver, I created a supplement stack around glutathione support and general inflammation. This is what I settled on:

• S-acetyl-L-glutathione, 200 mg per day
• NAC, 600 mg per day
• TMG, 4 g per day
• Turmeric, 1 g per day
• Grape seed extract
• Fish oil, specifically Nordic Naturals Complete Omega Xtra
• Elderberry extract, 500 mg per day

The liver part was mostly SAG plus NAC. NAC is a precursor for glutathione synthesis [7], and SAG was my attempt to get pure glutathione into my body without injections. TMG was there because I was interested in gains at the gym, and high doses of TMG are known for helping muscle growth [8]. Fish oil and turmeric were there for inflammation, and elderberry is more of a general-purpose antiviral I take to stay healthy.

Was this perfectly controlled science? NOPE! But I was going to do it all anyway and see what happened.

At this point, I had changed a bunch of these things at once: no alcohol, better diet, supplements, more health tracking, and a level of motivation I do not usually have.

To be fair: I started the supplements seven months after I stopped drinking, and my data show that my ALT and AST levels dropped to just-above normal levels when I started supplementation. With drinking, it’s usually understood that the liver completes its recovery in a few short months. For this reason, I’m inclined to believe that the supplements were the thing pushing me over the edge here, and into B.L.O.A.T. territory.

After running the stack for about three months, my numbers came back into the normal range. Nice! My ALT and AST were finally ‘normal’. They were at the higher end of normal, but still normal.

The problem was that normal was only my first goal. B.L.O.A.T. was still out there.

Going one step further

Once the oral stack got me back into a safer place, I doubled down on beating my old best test results.

I started by bringing back injectable glutathione. I went back to the exact same men’s clinic that I had visited years ago, paid to see a physician, and explained to her everything I’ve just shown you. While she thought I was a little obsessed over it, she said it couldn’t hurt to try. We settled on 600 mg weekly, slightly more than my OG prescription years ago. It was still painful, and still sucked to do every week, but it was just for a few weeks. I was doing this to give my liver a short-term boost to help it regenerate [6].

This was the differentiator. A short few weeks later, May 2026 came around.

The rematch

Here is the oldest test against the newest test, complete with the trends we talked about before:

I started chuckling at the new result. 2026 wins.

I know that not every single marker was better in 2026, but most were. A/G ratio was slightly worse, bilirubin rose a little, so if I were trying to fight against myself, there are places to do it. But if you look at the liver enzymes, they are ridiculous now. AST dropped from 147 to 25. ALT dropped from 71 to 17. Even compared with my seventeen-year-old baseline, AST is better and ALT is the same. That is insane to me.

What I think worked

If I had to rank the interventions, sobriety and glutathione tie for the top spot.

Sobriety is important because you can’t really expect your liver to heal if you are continually bombarding it with stress. Cutting alcohol created the perfect space for my liver to regenerate itself, and I’d recommend anyone do the same if they are in a similar situation.

However, I do think there is a lot to say about injectable glutathione. Injectable glutathione had the clearest impact on my liver enzymes by far. I want to slightly push back on the notion that sobriety is the #1 thing that is going to help someone’s liver get to its best functioning state of all time. While avoiding heavy drinking is an absolute must, my data showed that it only led to modest improvements. To actually get to best-functioning territory, the injectable glutathione pushed me over the edge. Take a look at the timeline of my biomarkers below. If sobriety really was the main thing driving this, we would have seen remarkable improvement in these enzymes. We didn’t. The most dramatic drop was actually not the time when I went sober, but the time that I started doing intramuscular glutathione. Keep in mind this was only for about 6 weeks. I think this is as good a testimonial as any, that these interventions do work.

With ALT especially, it's clear that the most significant drop in enzymes occurred when I was using intramuscular glutathione. The original drop from March 2025 onward was after 3 months of sobriety + 6 months of supplementation. Also - this is my new blood test tracking app, developed as yet another side project. You can read about it [here](https://sull1van.com/tracking-blood-tests) if interested.

As far as the supplements go, I don’t think any oral supplement is going to outsmart alcohol if heavy alcohol consumption is the main thing stressing your liver. But I do think oral supplements help, such as the SAG and grape seed extract. The research is promising [4][5], and my test results add more fuel to that fire. S-Acetyl-L-Glutathione is the only one I’d take orally. And after injecting pure glutathione, I learned that it costs only a little more than the supplement anyway. So for anyone who is considering one of the two, I’d say do whatever you are most comfortable with, but know that both will probably help you out in some capacity.

And that’s it

I started with a simple goal: get my liver back into range.

Then, because I was mad at myself for letting it fall out of range, I turned it into a competition against my own teenage bloodwork. And take that, teenage me! Twenty-five year old me is just as good, if not better.

As of May 2026, I have the best liver test results of my life. Maybe not every single marker, if you want to pick at it. But taken as a full picture, especially with AST and ALT, this is the best my liver has ever looked on paper.

B.L.O.A.T. achieved.

References

[0] Liver Function Tests: https://www.ncbi.nlm.nih.gov/books/NBK482489/

[1] Enhancing the Oral Bioavailability of Glutathione Using Innovative Analogue Approaches: https://pubmed.ncbi.nlm.nih.gov/40143049/

[2] Oral Administration of S-acetyl-glutathione: Impact on the Levels of Glutathione in Plasma and in Erythrocytes of Healthy Volunteers: https://www.graphyonline.com/archives/IJCND/2018/IJCND-134/

[3] Safety assessment of S-Acetyl Glutathione for use in foods and dietary supplements: https://www.sciencedirect.com/science/article/pii/S0278691525000468/

[4] Grape Seed Extract to Improve Liver Function in Patients with Nonalcoholic Fatty Liver Change: https://pmc.ncbi.nlm.nih.gov/articles/PMC3003214/

[5] Effects of grape seed extract supplementation on cardiovascular risk factors, liver enzymes: https://link.springer.com/article/10.1186/s12906-024-04477-3/

[6] Effect of injectable glutathione for treatment of liver disease: https://pmc.ncbi.nlm.nih.gov/articles/PMC5549431/

[7] How the body synthesises NAC into glutathione: https://pmc.ncbi.nlm.nih.gov/articles/PMC8234027/

[8] Baseline clinical study about TMG which shows it is promising for use in lifting: https://pmc.ncbi.nlm.nih.gov/articles/PMC3844502/

✕ click anywhere to dismiss

Over one year ago, I made the decision to start focusing on my bloodwork. I had just gotten the first comprehensive metabolic panel in years, and the results weren’t great. My liver enzymes ALT and AST147ASTNormal: 10–40 U/L71ALTNormal: 9–46 U/LSource: LabCorp, March 2025 were extremely high, in such an embarrassing way that it basically screamed this person drinks too much alcohol [1]. I realized that a bigger effort was needed if I was going to take charge of my health.

I started by looking at the data. Luckily for me, I’ve always had some sort of a focus on health, and I already had some two-dozen PDFs of blood test results dating back to 2018 in a folder on my computer. Every time I would get one, I’d save the results from the doctor as a PDF and drop it into that folder. While it was great to have, it was a mess to actually go through. I really wanted to make some comprehensive Excel spreadsheet with all of it, but assumed it would be too much work for one person to do.

My Plan

My plan was pretty simple, just:

1. Command Claude or Codex to delegate an army of subagents to process all my PDFs
2. File by file, comprehensively pull all numeric and non-numeric test results out
3. Create a master datasheet with all these results
4. Run a separate set of agents to check the first agents’ work, and report any inaccuracies they find.
5. Create a central dashboard to read and manage all my blood test results

It’s a lot. But since AI resources are virtually infinite, I was confident that it could be built.

How it went

Steps 1-4: Data conversion

This was honestly easier than I expected. Claude and Codex both did a great job dispatching agents and converting the blood test results into an Excel sheet. The cool part is the verification bots didn’t find any actual errors, they just flagged places where the wrong units were used, like an ALT test from 2018 in ng/dL while the 2020 ALT was in ug/L. I did a manual spot check across the whole dataset and it looked great.

Step 5: The Dashboard

With the data in some kind of usable format, this was the part where I had some fun. I started by asking Claude and Codex to both design a rudimentary visual for the labs. My idea was pretty simple. It would have a dual layout, card-based and list-based, where you pick between the two views and get either one card or one list item per blood test. They always show the most recent result and its unit, and they’re color coded green or red depending on whether the result falls outside the normal range.

The Frontend

Claude iterated on my design a few times, and the early prototypes were ugly. Inspired by the latest iOS update and wanting something a bit better than flat, I pushed the AI toward a glassmorphic UI. I’m really proud of how it turned out.

Card versus list UI view

The UI comes with many different modes, meant to highlight different points throughout the day. I wanted to simulate the UI of the Apple Weather app, because that is my ideal design for something that shows health data. It actually took a long time to achieve this UI, so I'm proud of the result.

**Left**: detail view showing most recent test, normal range, and units. **Right**: view when you scroll down. Shows all test history, and offers context menus which present additional options

The card view with daytime mode enabled. It was extremely hard to get the UI to this state, so honestly I'm blown away it all works.

Screenshot of Cmd+K. I don't have a better name for it, but it's triggered by Cmd+K on desktop, so

Once I finished, I spun up some more Claude subagents to do bug catching, and in the middle of that one of them suggested a new feature I really liked. That’s what you’re seeing in the last screenshot, actually. It’s a power menu that’s triggered by tapping the little flask on mobile or hitting Cmd+K on desktop. When you open it, you can instantly pull up test results, create an export PDF, or even add a new lab. It’s really powerful and designed to be my place to drop in when I need to locate a result as fast as possible.

That’s pretty much the entirety of the UI. I really loved designing it, especially the glass background. I was going for an imitation of Apple Weather, and I think we (me/Claude) nailed it.

The Backend

For backend storage I went with Google Firebase. The only real alternative people seem to be using nowadays is Supabase, and that has an awful free tier. Google’s looked better for the long haul. But since Google prices based on # reads, the data model I ended up with is honestly pretty funny.

Firebase gives you 50K reads/writes a day for free. After that it’s some absurdly low rate, like $0.03 / 100K reads (i think). The other great thing is the 1MB data record limit. 1 megabyte! 😂 That’s about 1 million characters. For reference, the Great Gatsby comes in around 240,000. My blood test results are never going to be as long as the Great Gatsby, so therefore I decided to store my entire database in one Firebase entry.

It’s safe to say that this setup will never cost me a dime. The only remaining question is what the data will look like when we send it to Firebase. I iterated with Claude, and we settled on the following schema for a master format:

  {
    "schemaVersion": 1,
    "tests": [
      {
        "name": "string",
        "unit": "",
        "range": "",
        "type": "",
        "points": [
          {
            "date": "YYYY-MM-DD",
            "rawResult": "",
            "lab": "",
            "sourceFile": "",
            "testResultId": "",
            "rawRange": "",
            "unit": "",
            "sourcePage": "",
            "reviewStatus": "",
            "verificationStatus": ""
          }
        ]
      }
    ]
  }

This schema looked pretty good. I sent it to firebase, but then realized I just stored private health data in plaintext, which is stupid. It definitely needed to be obfuscated. I thought about just encrypting the values themselves, and leaving the json keys themselves plaintext. But that is also pointless since the metadata (test name, test date) matters just as much. So I settled on a Base64 AES-GCM ciphertext of JSON.stringify(vaultData), which turns the whole thing into a string and encrypts all of it.

With that, it finally looked good. I wired up the Firebase integration and ran a one-time script to push all my data, and it worked on the first try. I’ve been using this site ever since then to track my test results.

The firestore database has essentially become my new source of truth. I still make regular exports so there’s the Firestore database and a Excel sheet running in parallel, giving me dual sources of truth. It’s a little extra effort, but I want to be protected in case something gets mangled in the database and my data becomes unusable.

Closing thoughts

This project was genuinely a lot of fun. I reach for the web app whenever I need to look up a blood test value, and whenever I bump into a new bug or feature idea I can just build it. Over time I’ve added a ton of functionality, and it’s become a really solid home for my data. It’s also easy to share it, making conversations with doctors and family much easier. I might publish this web app to Github if there is interest, but for right now it’s just a happy webapp that sits on my phone.

It also helps feed my obsession of actually trying to optimize these numbers. Since creating this dashboard, I actually did a whole project where I used it to help dial in my supplement routine and help out my liver. If you’re interested, feel free to take a look at that here.

References

[1] Per NIH, https://pmc.ncbi.nlm.nih.gov/articles/PMC4155359/: “In most types of liver damage, ALT is higher than AST. However, in alcohol-related liver injury, the enzyme AST is typically two to three times higher than ALT. An AST-to-ALT ratio greater than 2:1 strongly suggests alcohol is the cause.”

✕ click anywhere to dismiss

One month on the road with a self-driving car

In summer 2025, I had a pretty big goal. I wanted to drive across the entire western half of the country on a solo road trip, within the span of one month.

It took weeks to plan out the complete itinerary, which involved stops in over 23 states, and visiting with twelve different friends all across the country, plus working out of three different company headquarters locations.

But the coolest part of the trip (by FAR) was how I did it with the help of a Comma 3X.

I set it up in the 2-3 weeks before the trip, and got all the kinks worked out just in time. In the end, the Comma handled the vast majority of highway driving, and I only ended up doing that last 30-40% of each drive manually.

I’d love to show you how I did it!

What is a Comma 3X

So Comma.ai is the name of the company, and the 3X is a small, windshield-mounted computer. It takes control of the steering in most production cars. It runs their custom software called OpenPilot, which is a fully customizable user interface for the device.

At first, I saw that my Kia wasn’t supported, so I gave up my hopes of autonomous driving as soon as I got this car in 2024. A year later,though, I joined a Discord and found out that several other Kia owners had retrofitted this to their (newer) cars, so I could do it too. I just had to use this obscure fork of the software I found.

Some helpful background: Comma’s software is open source, which means the official version is only stock openpilot. There are plenty of community forks for different cars, different preferences, and edge cases that Comma decided not to support. To that end, a group known as Sunnypilot actually made their own fork of it, adding new features. And within that Sunnypilot Github repo, there’s this fork some Hyundai/Kia/Genesis (HKG) owners made. THAT’S the one that I use. So if you end up googling Comma openpilot and see your car isn’t supported, I highly recommend doing some more research. In my case, my car wasn’t supported, but all I had to do was join the Discord, learn what kind of harness I’d need, buy that one, and figure out who in the community had already forked the code.

Installing it

This is where it gets fun..

You’ll need to look up the Comma setup video specific to your car. I found that all HKG cars have similar installation instructions, so it was pretty easy to get mine put in. 5 minutes of looking through YouTube and you should be on your way.

The interesting thing about the Comma 3X is that it sits between your car and the camera. It intercepts the feed coming straight from the camera, decodes it, and uses it for autonomous control. It also has a genius passthrough mechanism where it will pass the camera footage straight through to the car if the Comma loses power. When driving, this would save me in the rare instances in which the Comma was annoying me and I pulled its power plug. Once you do that, the car defaults to its own controls without any issue. It’s a highly resilient setup.

Setting it up

First launch

By far the hardest thing to work on with a Comma is the software. It will install openpilot by default, which will fuck you over if you are like me and don’t have a supported car. Don’t do this! You will spend so much time trying to force a factory reset only to reinstall openpilot by accident.

On setup, you need to choose “Custom software”, and enter in a link. What link exactly? This is where it gets weird.

After a ton of research, I learned you can point the Comma at a website with a release image of the software you want. It’s nearly impossible to figure it out, though. I literally could not find the link for sunnypilot. But then I learned that Comma will actually do it for you, as they host Sunnypilot on their own domain. You just type in “sunnypilot” in the custom software URL box, and that’s it. For me, I’m using a branch of the sunnypilot GitHub repo, so I do “sunnypilot/ccnc-port”.

Getting the specifics right

There are a million variables to configure once your Comma fully boots. The Comma will try to identify your car’s fingerprint. Fingerprinting is essentially the Comma identifying what car you own, and what APIs it’s going to need to interact with to control your steering. By default, it should be able to fingerprint your car. If it can’t, you’re screwed. It will let you choose a car manually from the list, but the fact that it couldn’t figure it out automatically means it can’t communicate with your car properly, so you really can’t choose it manually. This happened to me a lot when I would try different branches of Sunnypilot, and led to me doing endless factory resets. For me, the only branch of Sunnypilot that can fingerprint my car is ccnc-port, so make sure you use the appropriate port for your car. You can search the Sunnypilot Discord for your make and model and see what others did to get it set up.

After fingerprinting, you should go into the settings menu and start configuring things to your liking. Using custom settings, you can have it display the current speed limit, your current speed, and many other cool metrics. I turn most of these on. One thing Comma does that I’ve never seen anywhere else is it will actually show you the speed of the car in front of you. I don’t know why car manufacturers don’t do this to begin with, since all modern ones are seeing this data with their ultrasound sensors. Either way, it’s great to have this on the Comma.

There are also a lot of configuration settings that will affect your driving. It comes with dozens of driving models, each one being about ~70MB a piece and containing a slightly modified version of a driving model based on who trained the data. Comma releases these regularly, as they have their own in-house datacenter that generates them. No model is ever the ‘best’ one, since all cars are different. You really just have to start driving and figure out which works well for you. For me, that meant choosing like ~10 models and just spending the entire day on the road testing them.

You’ll also want to look at additional things, like how you want lane changes and braking to work. The Comma won’t brake by default, and it won’t do lane changes by default, but it can if you want it to. I personally do automated lane change, but no braking, since the Comma is pretty shit at managing my speed.

Driver monitoring

By far the most ‘controversial’ aspect of the Comma is that it enables hands-free driving. To accomplish this without being reckless, Comma developed a driver monitoring system that runs at all time. The Comma 3X has a wide-angle camera which constantly watches the driver and runs a neural network to assess their attention. For example, if the driver uses a phone during driving, Comma will notice that and start beeping.

Since the software is completely open source, you might think you could change these settings. Turns out, you absolutely can. For anyone wondering, it’s two files, helpers.py and monitoring.py. I have done extensive amounts of testing with mine, and have arrived at a happy medium for monitoring: I do WANT the monitoring, since hands-free driving could lead to instances where I become inattentive to what’s on the road, but it shouldn’t be annoying. I wanted the system to notice real inattention without yelling at me every time I glanced at navigation or checked something quickly at a stop. So I ended up changing the code, but only slightly. You may want to do this too.

One thing I like about the attention tracking is that if Comma notices me being drowsy, it will make an extremely loud beep to wake me up. This did happen often when I would drive late at night, and I am very thankful for it. The monitoring you get with the Comma is honestly great peace of mind. You can also be assured that your face isn’t being recorded and uploaded anywhere, because you control the software, end to end. It’s great.

Beyond the basics: Asking for help

If you do try to get set up with this, there can be tons of issues. Driverless car technology is insanely complicated, and was extremely difficult to get set up even for someone like me. The great thing about buying a Comma 3X, though, is that you’re joining a huge community of car drivers just like you. Sunnypilot and Openpilot both have large Discord chats you can join, and a lot of the useful information is scattered across threads, branch names, harness recommendations, and people casually mentioning the one setting that makes everything work.

Taking it out West

The Comma ended up coming with me on the trip, as I was able to get everything set up just in time. I would turn it on during highway driving and usually not much else. It’s not very well trained for steep turns, and will often overshoot them and drive the car slightly to the inside of the lane. Highway driving worked perfectly though, and it did a great job of keeping me centered. The really cool thing is that they’ve trained it in all kinds of driving scenarios, so it can reliably steer during the day, at night, and even during rain.

As I went across the country, I started using it more as a copilot than something which completely replaced my driving. For example, knowing I’ll be going through a stretch of straight, level highway with no one else meant that I’d turn it on and then focus more on the scenery than the drive. Sometimes I’d turn it on while I’m going to change climate settings or otherwise occupied, using it as insurance that I didn’t swerve when I did whatever the task was. I would also use it while I was actively steering, almost to provide insurance that I’d stay on the road. If I ever got too close to the edge of a lane, Comma would gently steer me back to the center (This is in contrast to cars that only steer to block lane departure, as they don’t continue steering to KEEP you in the middle of your lane. Comma does).

In all, it was a lot of fun. I really enjoyed being out West and I think the comma was part of the reason the trip went so well. It takes a lot of the headache out of driving, and makes it more of an adventure.

Recently I’ve taken it upon myself to study for the FE exam. There’s a longer term career pivot in the works, but for now I’ve been focused on getting the best study material possible.

At first, I found the Kaplan/PPI study material. They have a lot of questions, but some of them were written poorly and also just incorrect. After paying $150 for a one month membership, I scraped all the questions I could find and ended up with a dataset of over 1000 questions. I don’t think they are as relevant to the exam, though, so I found another resource that I think is much better.

Enter This site. I found a reddit post from another engineer, who said he sourced as many FE questions as possible from around the web, and came up with this dataset. He posted it for free at https://journey2pe.com/. I took his question list and tried my best to parse out the questions vs. the answers, and then format it into a text-only database (which was tough, since it has images). I was able to use AI and spend about two weeks concocting my own website, which is more interactive than his. It’s his data, but a better way of visualizing it.

If you’re interested in studying, take a look. I built it for myself but it should be pretty easy for others to pick up.

https://sull1van.com/FEStudy

Have you ever wondered if there are ‘good’ and ‘bad’ chargers? Is there really a difference between a $5 brick you get at an airport and a $30 one you order direct from a name brand? As it turns out, cheaply made chargers can operate far differently from name-brand ones, and can sometimes be dangerous to use.

This work was done first by one Google engineer Ken Shirriff many years ago. It was at the top of Hacker News at the time, and has been reposted a couple of times. Ken made the case that there is absolutely a difference between the best and worst chargers, and that poorly built chargers have the ability to not only degrade battery life but cause other strange issues like touchscreen glitches.

Fast forward six years and it still looks like there is no real authoritative source that tests big batches of chargers like this. My friend Spencer joined along, being a mechanical engineer, and we started bouncing ideas back and forth until this project eventually came to life. He helped complete a lot of the work below - thank you, Spencer!

Here’s the synopsis of everything below:

• GaN chargers are among the best you can buy today, and the marketing around them is real.
• Cheap chargers can also perform well, so the price you pay for a charger doesn’t always correlate to its performance.
• Common brands like Apple generally do test favorably, but some smaller brands actually do better.
• There is an extreme amount of variation even within the third party brands. Some third-party produce both high-quality and extremely low-quality chargers, and you’d have no way of knowing without testing them.

What we tested

To get as much out of this project as possible, we tested a lot of chargers. Here’s the full list:

• Anker ASPD053a USB C
• Anker 637 Magnetic Charging Station (USB A port and USB C port)
• AOHI GaN 65W AOC-C019
• AOHI GaN 100W AOC-C018
• Amazon Kindle 9W PS57CP
• Dealworthy 20W PYSD-20PCQ-1CU
• SHARGEEK 35W Retro S035U
• Apple HomePod Mini A2305 (Same as Apple 20W Retail charger)
• Apple 30W Retail A2164
• Eero 15W C110011
• Anker 100W A121B GaN Flagship
• Anker Nano 45W A121D GaN Flagship
• Spigen PowerArc ArcStation Pro 30W GaN PE2013UJ
• Apple 96W Retail A2166
• VEBNER 65W PD65U55
• Govee 10W BI12T-050200-BdUU
• Counterfeit AT&T 10W ‘NA010050020’
• Counterfeit Chinese Wall Adapter 10.5W WP2UWR
• Counterfeit Samsung 10W EP-TA20JBE

There was also a Macbook Pro M1, an iPhone 17 Pro, and an Anker 621 MAG GO 5000mAh Battery thrown into the mix. These aren’t chargers, so I’ll keep them out of the list above, but they are included in the various results below.

Each charger was tested on several dimensions: Advertised power versus actual power, power quality, and overcurrent.

• Advertised power versus actual power tells you how well the charger can maintain its intended voltage while putting out maximum current.
• Power quality means the amount of noise that appears when viewing the charger with an oscilloscope. This noise comes in many forms, but the general idea is that the flatter, the less noise.
• Overcurrent shows how a charger behaves when a device tries to pull too much current. This is achieved by connecting a charger to an electronic load and having that load request more and more current until the device shuts off. It helps show how well-regulated a charger is, and if it is capable of protecting itself during load.

Testing Setup

Testing USB C chargers is a little complex. Modern chargers rely on the USB C Power Delivery (PD) standard, which generally specifies five voltages they can choose from: 5V, 9V, 12V, 15V, 20V. There are modern variants of this that complicate things even more, such as Programmable Power Supply (PPS) and Adjustable Voltage Supply (AVS), but I won’t get into those. For this project, we only looked at chargers according to the USB C PD spec. USB C PD also specifies that devices should be within +/- 5% of their advertised voltage in order to achieve certification (this is oversimplifying, but the general rule), so we followed that rule when rating them.

With this in mind, our setup consisted of a few different pieces:

1. An oscilloscope, specifically the Rigol DS1054Z
2. An electronic load
3. A USB C Power Delivery decoy board
4. A voltmeter
5. Plenty of electrical wires and extra USB cables

The process involved connecting the charger into a power source, and connecting the other end to a USB C Decoy Board, and connecting that decoy board to an electronic load. The oscilloscope probes tapped in right at the decoy board [A1].

The decoy board was needed because the USB C PD Spec needs the load device to choose a voltage. Otherwise the charger won’t send anything. In the chargers Ken tested, most often they’d just send 5V regardless, so this setup was a little more complicated.

A note about resistance

One of the things we learned during this project was that resistance could be artificially skewing the measurements. We did a couple of things to minimize this, so it shouldn’t be a big deal.

Rating system for Oscilloscope graphs

Noise and Switching Ripple

After conducting the oscilloscope tests, we rated each charger from 1 to 5, using lightning bolts as a visual. These ratings consider two separate factors: overall high-frequency noise (Vpp/spikes) and switching ripple.

• Noise includes the more chaotic high-frequency artifacts, large vertical spikes, and overall ‘fuzziness’ in the output. By measuring the overall peak-to-peak (Vpp) voltage, you can rank the true noise.

• Switching Ripple refers to the high-frequency peaks (e.g., 50KHz, 100KHz) caused by the charger’s internal oscillator continually charging and discharging. This differs from Ken’s methodology, as he used the word Ripple specifically to mean the 120Hz line-frequency fluctuations caused by household AC power.

• Most of the chargers look incredibly dense and ‘thick’ compared to the ones Ken examined. I believe this is because my scope has a memory depth of 24 million points, while the most I saw on Ken’s graphs were around ~50,000 equivalent points of memory depth. I believe the inclusion of this raw high-frequency data at every zoom level is the most honest way to display these chargers.

For noise, the bolt ratings loosely map to measured peak-to-peak voltage noise on the oscilloscope:

• 1 bolt: over ~250mV
• 2 bolts: under ~250mV
• 3 bolts: under ~160mV
• 4 bolts: under ~100mV
• 5 bolts: under ~80mV

These are just general guidelines. I used my judgment for each charger based on everything we could see, including the shape of the waveform and the FFT spectrum, not just the raw mV reading.

The Results

Scroll through the cards and click any one to expand the full oscilloscope graphs. Each card shows the worst graph that we could reproduce for that charger, assuming typical power usage.

A few more details on noise and ripple

They look confusing at first glance, but you don’t need to be an electrical engineer to understand these next graphs.

• Yellow Line (Voltage Trace): This represents the actual voltage being delivered to your device. The horizontal axis represents time, and the vertical axis represents voltage. In a perfect world, this would be a single, razor-thin, perfectly flat horizontal line. Any deviation from this is an imperfection.
• Purple Line (FFT Spectrum): You’ll see this purple mountain range near the bottom of many graphs. This is a Fast Fourier Transform (FFT). It takes all the imperfections from the yellow line and sorts them by frequency. A large spike here would mean there is a significant amount of energy at that specific frequency.
• Noise vs. Switching Ripple: This describes the imperfections on these graphs. It’s important to look at them separately:
  • Switching Ripple is the presence of a main, high-frequency spike (and its harmonics) on the FFT. This is caused by the natural charging and discharging cycle of the charger’s internal circuit. AC Chargers will switch on and off many times per second to effectively create something that is DC current. For that reason, if there’s a significant peak at 90KHz and nowhere else, for instance, we’d say the charger has its primary switching ripple at 90KHz. If the charger has peaks at 90KHz and 180KHz, that is the switching ripple and its harmonic at 180KHz.

• Noise includes any imperfections on the voltage graph measured in peak-to-peak voltage (Vpp). There are many different kinds of noise, including high-frequency vertical spikes, fuzzy blur, and general jitter. Switching ripple is one component of this.

• ‘Fuzzy’ Blur: If the yellow line looks incredibly thick or “fuzzy,” it means the charger is constantly shifting its switching speed back and forth (frequency dithering) to dynamically smear its noise.

• Good vs. Bad: A high-quality charger produces a thin, flat yellow line with minimal action on the purple FFT graph (ideally just one sharp needle representing its switching ripple). A poor-quality charger will have a thick, fuzzy yellow line (lots of jitter/noise), obvious rolling waves (line ripple), and many aggressive, evenly spaced harmonic spikes on the purple FFT graph (severe switching ripple).

Oscilloscope Graphs

Macbook Pro M1

Overall rating: ⚡︎⚡︎⚡︎⚡︎⚡︎

Expand

AOHI GaN 65W Charger

Overall rating: ⚡︎⚡︎⚡︎⚡︎⚡︎

Expand

Anker 621 MAG GO 5000mAh Battery

Overall rating: ⚡︎⚡︎⚡︎⚡︎⚡︎

Expand

Spigen PowerArc ArcStation Pro

Overall rating: ⚡︎⚡︎⚡︎⚡︎⚡︎

Expand

Eero 15W Charger

Overall rating: ⚡︎⚡︎⚡︎⚡︎

Expand

AOHI GaN 100W Charger

Overall rating: ⚡︎⚡︎⚡︎⚡︎⚡︎

Expand

Anker 637 Magnetic Charging Station 2 - USB C Port

Overall rating: ⚡︎⚡︎⚡︎⚡︎⚡︎

Expand

Counterfeit Chinese Wall Charger (USB A Only)

Overall rating: ⚡︎

Expand

Apple 96W Macbook Charger

Overall rating: ⚡︎⚡︎⚡︎

Expand

Apple 30W Retail Charger

Overall rating: ⚡︎⚡︎⚡︎⚡︎

Expand

Anker ASPD053a

Overall rating: ⚡︎⚡︎⚡︎

Expand

iPhone 17 pro

Overall rating: ⚡︎⚡︎⚡︎

Expand

Shargeek Macintosh Charger

Overall rating: ⚡︎⚡︎⚡︎

Expand

Govee Power Supply

Overall rating: ⚡︎⚡︎

Expand

Dealworthy Charger

Overall rating: ⚡︎

Expand

Counterfeit AT&T

Overall rating: ⚡︎

Expand

Anker A121B 100W Charger

Overall rating: ⚡︎⚡︎

Expand

Amazon Kindle Charger

Overall rating: ⚡︎

Expand

Apple HomePod Mini Charger

Overall rating: ⚡︎

Expand

Vebner GaN 65W Charger

Overall rating: ⚡︎

Expand

Anker A121D 45W iPhone 17 charger

Overall rating: ⚡︎

Expand

Counterfeit Samsung

Overall rating: ⚡︎

Expand

Oscilloscope Graphs - Results Table

The DC items in here, like the Macbook Pro, are colored in grey to avoid conflating them with the AC chargers. They’re just included here for reference.

Expected/Actual Output

The Actual Voltage column is highlighted green if the measured voltage is within ±5% of the nominal output voltage (the USB-C PD spec limit) and red if it falls outside that window in either direction. For example, a 5V output must stay between 4.75V and 5.25V.

Expected/Actual Output Table

The DC items in here, like the Macbook Pro, are colored in grey to avoid conflating them with the AC chargers. They’re just included here for reference.

Overcurrent Testing

The last piece of Ken Shirriff’s testing we wanted to replicate was overcurrent. Naturally, since devices get to decide how much current they want to pull, it’s up to the charger to make sure the device doesn’t pull too much current. Chargers generally have two strategies to deal with this: 1) shut off completely, or 2) reduce the output voltage. Option 1 is preferred in most cases. Option 2 is really just a consequence of not doing option 1. There is a third scenario, where the charger can somehow increase the current at the same voltage, but this would probably mean overheating it.

The ideal value here is having a voltage equal to 5V or 9V. You don’t want 3 or 4 volts, or 7 or 8 volts, because that suggests the charger allowed more current out than it should have.

We don’t have overcurrent results for all chargers in the table below. This will be updated as more test results become available.

Overcurrent Results Table

A mixed bag of results from overcurrent. The Kindle is the most disappointing one in my opinion, because it shut off so quickly at just 0.1 A above its rated current, and somehow its voltage still dropped. However, the worst offenders are from the Apple 96W, Spigen PowerArc, Eero, and the counterfeit Chinese wall charger. All had voltage drops that suggest they go below what would be considered USB standards.

A new update: I re-ran the tests one more time with a brand new multimeter and updated the table. Upon reviewing, the Spigen, Apple 96W, Counterfeit Chinese wall charger, AOHI GaN 65W, and Amazon Kindle, and Eero are all starting to break past the +/- 5% voltage threshold for USB certification. I re-tested them under the oscilloscope, and found some interesting results: while most of these chargers offered relatively focused voltage traces, usually ranging by about 0.1V, the counterfeit Samsung was the one outlier. This charger had such a noisy voltage trace that, near shutoff, it oscillated between 3.7 V, 0, and 8.5 V. This dangerous behavior was truly unique and not present in *any of the other chargers I tested. When your phone or laptop attempts to pull too much current, this charger will start sending massive shockwaves of voltage.*

—

Conclusion

Wow, that was a lot. Evaluating chargers seemed quite difficult at first, but this whole project quickly turned into something exciting. What I liked most about this is that you don’t need to be an electrical engineer to see the difference between good and bad chargers. The oscilloscope really does make it obvious. Things get confusing when you go past the ‘what’ and try to start understanding the why. Why certain peaks appear in the FFT, why some scope traces look as bad as they do, or why a charger measured fine at 1A but looked terrible at 2.5A. I don’t think Spencer or myself have great answers to these.

The practical takeaway, in my opinion, is that most name-brand chargers of today are unlikely to damage your phone battery. Since modern devices filter noise internally, and the majority of these chargers stayed within that USB spec of ±5% voltage tolerance, they’re unlikely to be a problem. Just avoid excessively cheap, no-name chargers and you should be fine.

I hope you enjoyed reading. This really was a pleasure to put together. Car chargers are next on my list, and I suspect that may be a story of its own. If you have any suggestions to make, corrections to add, or just want to share something you found interesting, feel free to reach out and let me know. Contact is on the About page.

-Sean

Last thing: a huge shout out to both Ken and Spencer! I could not have done this without either of them.

Sean Sullivan
A Dozen Chargers in the lab

Appendix

[A1] Tip: If you want to do testing of your own, always use the spring ground clip. Using the default alligator ground clip will lead to extremely high noise ratings.

For years, I had this website as an email domain, using it with iCloud so I could email people from @sull1van.com. I realized that I made a lot of personal content which should be published somewhere, so I turned this into a real site. I’m an industrial engineer currently living in the United States, with passions for engineering and health. Most of the things I publish will be related to those two things.

Other info

This blog is built with Jekyll and focuses on content over presentation. I tried using blogging services but they were all pretty bad, so I vibe coded my own.

Stay tuned for more posts!