The Mission · Vision

Pointing AI at the diseases that take who we are.

The belief, stated plainly: aimed with care, AI can help us understand and one day predict the diseases that take who we are. Not from a laptop alone, and not overnight — but by getting the checkable pieces exactly right, then building outward, year after year.

← The Mission
Why AI, why now

Aim it where there’s an answer key.

My thesis has always been the same: the hard problems don’t need better AI — they need someone who knows how to aim it. This is me aiming it at the hardest problems we face.

Something real has shifted. AI can now contribute to the work of discovery — proposing, modeling, testing, and reasoning over evidence, not just summarizing what’s already known. A careful, methodical effort aimed at the right question may now reach places that previously required a full lab and a decade — though that claim only holds when the work stays grounded and checkable.

The discipline is knowing where to point it. AI earns trust in exactly one place: where the work can be checked — against a measurement, a held-out outcome, an answer key that already exists. Aim it there and it is a genuine force multiplier. Aim it anywhere else and you get confident-sounding answers with no way to know they’re wrong, which in medicine is worse than nothing.

Neurodegeneration has those answer keys. Which mutations are pathogenic is measured in the lab, and much of that data is public. How fast a person declines is measured over time, against what actually happens. Because those answers are checkable, a careful effort can make real, accountable contributions to them now — not someday. That is the whole bet: aim high, but only where the evidence can keep you honest.

Why neurodegeneration

Personal first. Then universal.

This started close to home — the full story is on the motivation page.

But the choice of domain isn’t only personal. Neurodegenerative diseases are among the most feared and least predictable conditions we know — precisely because they take the things that make a person themselves: memory, speech, movement, will. And one of the field’s weakest points is the thing families need most: honest, calibrated, individual prognosis. Not a population average. Not a brave-faced reassurance. An answer to “what will this look like, for us, and when” — stated with real uncertainty instead of false confidence.

That gap is the opening. It matters enormously, and a forecast about decline is graded by what actually happens — so it is exactly the kind of problem worth aiming everything at.

The roadmap

Years, not weeks.

This is a long arc, and I’d rather show you the whole of it than oversell the pace. Each stage opens only when the one before it is proven; some wait on data access I don’t yet control. That is the honest shape of an ambitious plan.

Now · Alzheimer’s & ALS

The two I started for

The molecular track — which mutations drive the disease, checked against public lab data — is already running. The clinical track, calibrated forecasting of how ALS and Alzheimer’s progress, is what I started this for. It opens the moment data access lands.

Next · A repeatable method

“How will this progress?”

The question families ask most has its answer built into time itself: what eventually happens. The aim is a calibrated way to say “here is how this is likely to go, and here is how sure we can be” — one disciplined method, proven once and then repeated.

Later · The wider family

Beyond the first two

If the shape holds, it should generalize across the neurodegeneration family — Parkinson’s and others — and across modalities: molecular variant effects, clinical trajectories, the speech signal. Multi-disease, multi-modality. Same discipline each time.

Here is the honest ledger. The molecular work uses public data and is running today. The clinical and speech tracks are gated on data-access approvals, and won’t begin until those land. The wider-family expansion is years out and depends on the earlier stages actually working. A solo with AI has real limits: I can build, calibrate, and verify; I can’t run a clinic or a wet lab. None of that dims the goal — it just sets the pace.

So the bet isn’t a single leap. It’s compounding: small, verifiable contributions, accumulated honestly over years, each one checkable and each one building on the last. Slow is fine. Wrong-but-confident is not.

The discipline

Ambition without overclaim.

The reason to trust the ambition is the ethic underneath it. The same rules apply to every stage of the roadmap, and they don’t bend for a more exciting headline.

The rules that make the ambition trustworthy
  • Calibrated uncertainty. Every prediction comes with honest error bars. A wide, truthful range beats a narrow, confident one.
  • Never a confidently wrong number. If the model doesn’t know, it says so. The failure mode I care most about avoiding is sounding sure and being wrong.
  • Every claim verifier-grounded. Nothing gets stated that can’t be checked against a measurement or a held-out outcome. No answer key, no claim.
  • Shown in the open. The results and the caveats both, as they happen — in the build log, with the data stewardship rules laid out plainly.

That’s the whole posture: aim high, but only at things you can check, and keep the checking visible. The ambition is the point. The discipline is what earns the right to have it.

Where this is going

What this is aiming at.

This started close to home — the full story is on the motivation page. What it aims at is a world where families facing these conditions get honest foresight — a calibrated sense of what’s coming, and roughly when — while there’s still time to use it.

Not a cure. Not a miracle. Just the thing I kept wishing for, made real and checkable, one verifiable piece at a time.

Read the mission → Why this exists →