Service · from $7,500
Screening in-house, without spending a year finding the traps.
We build the screening stack we would want to inherit: calibrated, reproducible, and structurally unable to quietly return fabricated data. Then we hand it over with the runbook.
From $7,500, scoped on the call. Typically four to eight weeks depending on scale.
Who this is for
Teams that have decided screening is core enough to own rather than outsource — and would rather not rediscover, one painful quarter at a time, every failure mode the field already knows about.
It is also for teams who already have a pipeline that has stopped being trustworthy: results that do not reproduce, thresholds nobody can justify, a scoring path that silently falls back to something other than the real model.
What we actually do
- Structure and affinity. Co-folding structure prediction with calibrated binary and affinity heads, plus a docking prefilter where it earns its keep — and not where it does not.
- Orchestration. Batch scoring that survives a job dying halfway: per-compound checkpointing, resumable runs, and wall-clock guards so one pathological input cannot eat the queue.
- Provenance. Every score traceable to its input, its engine version and its run. This is what makes a result defensible a year later when someone asks where it came from.
- No mock data, enforced. A scorer that cannot reach its backend must fail loudly, never substitute a plausible number. We enforce this in code because good intentions do not survive a deadline.
- Calibration and thresholds. Derived from your measured data with confidence intervals, not inherited from a paper about a different target.
- Reports. Versioned, reproducible, and readable by someone who was not in the room.
What you get
- A running pipeline on your infrastructure — or on hardware we specced for it.
- The calibration harness, so you can re-derive thresholds as your data grows.
- The adversarial gate suite, so new targets get checked before they are trusted.
- A runbook written for the person who inherits this after you.
If you would rather not run hardware at all, we can run the campaigns for you instead.
What we will tell you for free
Sometimes the right answer is not to build it. If you are screening a few hundred compounds a year against one well-characterised target, a pipeline is overhead you will resent. We would rather say that on the call than take the engagement and have you find out later.
Tell us what you want to screen, and at what scale.
We will tell you what it takes to do it properly, and whether you should build it at all.