Gene-Level AI Health Management — A Few Hundred Yuan of Genetic Data + AI = Personal Physician Parity

“Nutrition management and disease prevention precise down to the gene” — modern medicine can already do this. Ordinary people just can’t afford it.

I used to run the personal physician business at Xingshulin, building bespoke physician services for entrepreneurs and corporate executives. I know exactly how expensive that used to be. Today is about driving the price down to where ordinary people can afford it.

1. Genome sequencing is the most worthwhile test I’ve ever had

Over a decade ago I had full genome sequencing done at WeGene. A 20-plus-megabyte document, close to 1.2 million rows — one locus per row, A/G/T/C, the body’s source code.

Why is it the most worthwhile? Because your genome never changes, but the science keeps advancing. Every other test gives you one result today and a different one tomorrow; you have to keep repeating it. Sequence once and it serves you for life, letting you understand your own body better as the science improves.

A few hundred yuan gets it done. Those several-thousand-to-ten-thousand-yuan “full interpretation” packages on the market aren’t necessary — the extra money is buying you the “interpretation” part, and interpretation is exactly what AI can now do instead.

On the WeGene site: My Genome → Raw Data → download the “core data” (not the extended data, which is 10-million-plus inferred rows and isn’t accurate). With other providers, just ask support for the “raw data.”

2. Don’t just hand it to the AI — you need an allowlist

Dumping 1.2 million rows straight into an AI is useless — the vast majority of those loci have nothing to do with your health.

What you do instead: build an allowlist of gene sequences confirmed by clinical trials, drawn from top-tier global medical databases like PubMed — roughly 10,000-plus entries. The expression of these loci already has strong established correlation with human disease and pharmacology. These are credible, not fortune-telling.

Apply for a PubMed API key at NCBI (National Library of Medicine) and the AI can pull papers several hundred times per minute for source retrieval and evidence-building.

3. What the AI did for me goes beyond reading my genome

Before my checkup late last year, I had this AI system run through everything first. Here’s what it did.

I went to the Meinian Onehealth site and had the AI scrape the prices and test items for every men’s package, then combined that with my genetic data and my medical history to build me a tailored screening plan.

• A custom screening plan: core mandatory tests plus high-risk specialty panels, nothing missed. Last year I was diagnosed with cubital tunnel syndrome (nerve compression from long hours at a computer), so this year it had me redo EMG and nerve conduction velocity; hypercoagulability, metabolic panels, imaging — all covered
• An execution plan: it recommended Meinian’s “Deluxe Premium Men’s Checkup Package” at ¥2,749 to cover most items, listed the remaining few to do at a Class-A tertiary hospital, and laid out the appointment process
• Pharmacogenomics: which drugs I metabolize slowly, which ones won’t work well on me — precise input for my attending physician’s prescribing
• Exercise guidance: I “warm up fast but recover slowly” — my heart rate stays elevated after stopping high-intensity exercise, which badly disrupts the first half of my night’s sleep. So I moved football from 10 p.m. to 6 or 7 in the afternoon
• Supplements: dosages calibrated to my weight and my age in my forties. For example, I have a B12 transport issue, which is part of what caused last year’s cubital tunnel syndrome, so I now take mecobalamin daily

Every single conclusion links back to the original paper. This isn’t the AI guessing. And because most of the studies use European and American samples, the system also calibrates against Chinese population databases.

4. A few specific risk loci (from my own report)

The first thing that jumped out of the report was non-alcoholic fatty liver disease and liver fibrosis — it runs in my family, my father is a leading cardiothoracic surgeon in China, and he drinks pretty hard. I like to joke that I’m a “Shandong-Liaoning hybrid,” so I can hold my liquor.

But the report was blunt: my risk of cirrhosis and liver cancer from alcohol intake is 2.44x.

Being able to drink ≠ your liver tolerating it. “Able to drink” is the speed at which you metabolize alcohol; “tolerance” is how much damage your liver can absorb. The two are not proportional. Metabolizing fast doesn’t mean your liver escapes harm. Without a genetic report telling you this, you would never know.

The second was systemic sclerosis / Raynaud’s phenomenon. I love football, and when I play in indoor domes in the winter my fingers and ears go numb and cold unusually easily. After seeing the report, winter football means gloves and ear covers, no exceptions.

The third relates to thrombosis risk. I ran a guesthouse in Yunnan for four years, and years at high altitude left my red blood cell and platelet counts a few percentage points above normal — add family history, and high blood lipids become something I have to watch closely. I’ve also stopped the creatine I used to take before exercise, and I top up potassium and magnesium in the afternoon.

5. The healthcare parity AI brings

Producing a health management report this precise used to require spending millions of dollars on a medical team (think Project Blueprint and the like).

Now: a few hundred yuan for one genome test + AI analysis + PubMed grounding + Chinese population database correction = a report of equivalent depth.

This is the healthcare parity AI brings — work that used to take a medical team costing millions of dollars, now driven down to a few hundred yuan.

We’ve been running this for a year with a group of executives and founders, using a “physician + health coach” model. The genome doesn’t change, so the report keeps updating as medicine advances — you only have to do the test once.

Source: 基因级AI健康管理.mp4 · ASR model gemini-2.5-pro · full text of the 12-minute original recording

[00:00] Hello, welcome back to Pinpin’s channel. Today let’s talk about some applications of AI and healthcare. As you know, I used to be at Xingshulin running the personal physician business. We built customized personal physician services for entrepreneurs and for senior executives at companies. And a very important point in that work is this — modern medicine can already do precise, gene-level nutritional management and disease prevention. So how did we pull that off? AI is involved in a huge way. Let me decode it for you today.

[00:33] So what you can all see here is a real genetic report of mine. Where did this report come from? About ten-odd years ago, I’d already had my full genome sequenced at WeGene. So what is genome sequencing? You know DNA is the code of your genes — everyone’s code is there from birth, and it doesn’t change. That code actually holds a lot of factors that are especially tied to disease and to an individual’s physiological state, and as science has advanced, more and more of those factors have been confirmed by clinical studies in recent years.

[01:09] Personally, I think genome sequencing is the most worthwhile of all the tests we get done. Because to this day it’s still a fairly cheap price — a few hundred yuan can get you a sequence like this. There’s really no need to do the tens-of-thousands-of-yuan kind, because the extra part is just a detailed interpretation of the sequencing for you — and that part can be completely replaced by AI. Why is genome sequencing the most worthwhile test? Because with other tests, I check today and tomorrow the result has changed, and I may have to go test again. But with sequencing, basically once you’ve tested in your life, you can keep learning about your own body’s situation as science advances.

[01:53] You can all see the data from my sequencing — it’s a huge notepad document, more than 20 megabytes in total, close to 1.2 million lines inside. Every line has a specific locus and what the gene expresses — anyone who took high-school biology knows it’s A or G, T or C, right. So that’s our body’s code. And how do you get this report? Wherever you are, WeGene or 23Mofang, any of these sequencing places — take WeGene as the example: go to the official site, “Me” → “My Genes” → “Raw Data” → “Download,” and inside there’s a “core data” and an “extended data” download. Downloading the core data is enough, because that’s the real lab data. The extended data has over 10 million lines, and it’s inferred, so it isn’t as accurate.

[02:53] If you’re with another genetic testing company, you can go to their support and ask for what’s called the “raw data,” because that’s what we use for the analysis. So how do we actually do this? In the process of building it, we pulled from the world’s top medical sites — including PubMed — and from all the known databases linking human genes to disease and to drugs, and we put together a whitelist of somewhere over ten thousand sequences. Meaning these are genes whose expression has been confirmed by clinical studies to correlate fairly strongly with all kinds of human disease and all kinds of drug responses, and that are referenceable and trustworthy — not the fortune-telling kind of testing.

[03:41] Once we have data like that, we compare it against the individual’s data and run a detailed analysis with an AI engine. How does an individual go do this? This site, NCBI (National Library of Medicine) — you can apply there for a PubMed API key. And that key lets your AI pull paper information hundreds of times a minute, so it can do real source retrieval and evidence-checking across your whole analysis.

[04:17] Once we have that whitelisted gene data, the next thing we actually do is hand the user’s physical-exam report to the same AI. For instance, before my checkup at the end of last year I used this system of ours. It put together a checkup plan for me — China’s checkup providers are pretty good these days, so I went with Meinian. I went to their official site and had AI scrape all the men’s package prices and all the test items. Then I could have the AI tailor this year’s checkup for me based on my previous exam reports and my genetic situation.

[04:55] What plan did it build for me? It lists out the core must-do items. Because last year I had cubital tunnel syndrome — from long-term computer use — and I went to Beijing Hospital specifically for it. So because of that clinical history, this year it had me re-check EMG and nerve conduction velocity. Plus some of the high-risk items from my genes, blood hypercoagulability and a metabolic panel, and the key imaging studies — it listed out every item I should test. Then it gave me an execution plan: going with the annual “Deluxe Premium Men’s Checkup Package” at 2,749 yuan covers most of the items; but a few of the deeper tests can only be done at a top-tier hospital, and it listed those out too. And afterwards I did book an appointment at Beijing Hospital and got that set of tests done.

[06:01] It also arranges the whole flow of the tests — where to book, how to book, how to get the tests scheduled on the same day. And once this year’s tests were done, we fed them into our AI system — basically checkup report + genetic data = my health management report.

[06:19] In here you can see it first lists out the high-risk points. The first one I think is pretty accurate, and that’s “non-alcoholic fatty liver disease and liver fibrosis.” That condition actually runs in my family — both my parents are doctors, my father is one of China’s leading cardiothoracic surgeons, and he drinks pretty hard normally. I often joke that I’m “half Shandong, half Liaoning,” so I can hold my liquor pretty well. But that doesn’t mean my liver can tolerate alcohol — how much you can drink and how much your liver tolerates alcohol are not proportional. I metabolize alcohol fast, but my risk of cirrhosis and liver cancer from alcohol intake is 2.44 times higher, so I absolutely need to stay off alcohol. Don’t be like the people who say “I can drink that much, I don’t need to quit,” right — it really does come down to genetics.

[07:40] The second one is about “systemic sclerosis,” including what it calls “Raynaud’s phenomenon.” I like playing football, and in winter I go play on the indoor domed pitches — really cold pitches. And I noticed that my fingers and my ears very easily go into a state that’s very cold and also numb. So thanks to a report like this, I paid particular attention to wearing gloves and ear protection the next time I play football in winter.

[08:08] Same for bone spurs, malignant tumors and thrombosis risk — because I ran a guesthouse in Yunnan for four years, and living long-term at high altitude by itself pushes my red blood cell and platelet counts a few percent above a normal person’s, and that actually brings thrombosis risk with it. Especially when blood lipids are high, and on top of that thrombosis runs in my family too. So going through this, I pay particular attention to those points, including nutritional precautions, including daily lifestyle precautions. Sudden cardiac death and cardiac issues too. One point in there is that I used to take “creatine” sometimes before exercising, and I’ve stopped all that now, and I also supplement potassium and magnesium in the afternoons.

[09:01] The report also lets me see “pharmacogenomics” — meaning which drugs I metabolize slowly and need to watch out for, and which drugs don’t have a particularly strong effect on me and what kind of drug could be substituted. All of that is a very precise reference I can give my attending physician when I do get sick.

[09:22] And then there’s my athletic performance — I’m “quick to warm up but slow to recover.” I don’t need much of a warm-up to get into a high-heart-rate, high-intensity state, but my heart rate recovers very slowly. What does that mean? When I stop high-intensity exercise, my heart rate stays higher than my normal rate. And that actually affects sleep. Like, I used to play football at night, finishing at 10, and then for the first half of the night my heart rate was above my resting rate, which really hurt my sleep. So now when I schedule exercise I move it up to 6 or 7 o’clock, or an even earlier slot, instead of doing it at night. This is all individual, by the way — this isn’t universal advice.

[10:09] Including nutrient absorption. The one in here I think matters more is “folate,” and actually most people have this folate conversion issue. What’s unusual in my case is “vitamin B12.” I mentioned my cubital tunnel syndrome last year — that’s nerve damage, and on top of that my B12 absorption and transport are impaired. So after that came up in my tests last year, I specifically started taking a vitamin B12 supplement every day, mecobalamin. And you’ll see in here that because a lot of the data comes from studies in Europe and the US, there are differences between ethnic groups, so my data also gets calibrated for the Chinese population through our database.

[10:51] And finally you can also see that every theoretical source and reference in here is checkable — you can even search up the papers it cites directly. And once a plan like this is done, it also gives me research on my day-to-day supplement needs — including my current weight, my age, I’m in my forties, right, so what dosage of which nutrients I should be taking.

[11:21] So through this kind of detailed gene-level screening and the whole health plan, we use a “doctor + health coach” model to give executives and founders really good health protection. This is the system I built back then, and we’ve been using it at scale since last year. A system like this really is the medical parity AI brings. To do this kind of deep research — it used to take, and honestly still takes, something like “Project Blueprint,” spending over a million dollars on a medical team to do the analysis and the customization, before you could pull off a precise plan like this.