We build the design layer for multi-specific biologics.

QAI Labs Ltd. is an early-stage company — a Delaware corporation headquartered in Sugar Land, Texas — building NOVA-3, a computer-aided design suite for multi-specific antibodies. Our thesis is simple: the valuable molecules are the hard ones, and the hard ones have to be designed, not screened. We are a platform company, not a drug company — NOVA-3 is the product; our two internal programs (P1, P2) exist to prove the platform on our own dime before partners trust it on theirs. They are evidence, not the business. This page is the plain-English version of what we're building and why; the Platform and Benchmarks pages carry the technical detail.

Four bets we're making.

None of these are guarantees. They are the reasons we think this company should exist now rather than later.

The value moved to multi-specifics — and the tooling just caught up.

Single-target antibodies are commoditizing. The molecules pharma now pays for — trispecific T-cell engagers, conditional costimulators, half-life-extended multispecifics — are combinatorial, and bispecific/multispecific approvals have accelerated sharply over the last few years. The specific why-now: the open-weight foundation models we wrap (Boltz-2, Chai-1) only became commercially usable in roughly the last 18 months, so a composition-first design layer on top of them couldn't have been built two years ago. Folding models predict a structure; they don't compose a multi-arm molecule from validated parts. We compose.

Designed beats screened.

Brute-force screening is slow and expensive. Starting from a curated, schema-typed library of validated modules and ranking in silico turns a months-long hunt into a short design loop — our target is sequence to in-vitro PoC in 8–16 weeks.

Pharma buys evidence, not demos.

Diligence teams need to reproduce a result and trust the invention date. Every NOVA-3 candidate ships with an Evidence Bundle — a SHA-256 root anchored to a public timestamp, with the commands to reproduce the claim. That is a procurement feature, not a science-fair trick.

Quantum, stated honestly.

We apply quantum methods where they plausibly help — active-space corrections on the hardest, strongly-correlated chemistry — validated on simulators today, with hardware access in progress; we report the hardware fraction of every job. We make no claim of demonstrated quantum advantage in ground-state chemistry. The restraint is deliberate.

Where we are today.

An honest snapshot. Everything below is live or in active preparation; nothing here is a projection. Forward-looking work is in the timeline further down, marked as planned.

Live now

Platform & public demo

The five-layer NOVA-3 stack, plus a no-login demo anyone can run in the browser — real ESMFold structure prediction and an interactive design Playground.

Live now

Module Library

40+ validated VH/VL sequences across 3 active-program targets, plus linkers, hinges, and effectors — each module provenance-stamped and schema-typed.

Designed · in-licensed

Two active programs · in-licensed IP

P1 EpCAM × CD3 × 4-1BB TCE (IMT030122, in-silico PoC), P2 EpCAM × 4-1BB × HSA (IMB030202, in-vitro PoC). Program IP in-licensed (Schedule B); chain-of-title in progress. Wet-lab validation is scheduled, not yet complete.

Live now

Evidence Bundles

Cryptographic provenance on every result — a SHA-256 root committed to a public timestamp via OpenTimestamps, with reproduction commands attached. Verifiable from the Predict page.

Where we're headed.

Shipped milestones are marked done; everything dated forward is planned and labelled as such. We update this list as reality changes it.

DONE · platform
NOVA-3 platform built — five-layer stack, Module Library, Evidence Bundles.
DONE · product
Public no-login demo shipped — live structure prediction and the design Playground.
DONE · IP
Two active programs designed; program IP in-licensed (Schedule B), chain-of-title in progress.
PLANNED · 2026
Wet-lab validation of the lead candidates — SPR/BLI affinity, ADCC, and the P2 in-vivo study.
OPEN · onboarding design partners
First external partnership on a partner's target — onboarding a first design-partner cohort, capped at 3–5 so each gets real attention. Sequencing is deliberate: prove the platform on our own programs, then run it on partners' targets.
PLANNED · Q3 2026
Platform preprint on bioRxiv, with a benchmark suite against published baselines.

For partners.

If you run pharma BD or a partner lab, we'd like to hear from you. The first conversation usually starts with a program walkthrough and an Evidence Bundle review.

QAI Labs Ltd. · Sugar Land, Texas · partner@qailabs.co