Perceptic, a drug-development startup founded by three engineers who built Palantir's Artificial Intelligence Platform (AIP), exited stealth on May 26 with a $12 million seed round led by Accel, with Air Street Capital and Elder Gull also participating. The pitch is deliberately narrow: not another molecule-predicting model, but the "operating system" that wires existing AI tools into the siloed data and decisions that actually run a pharma pipeline.
The team and the thesis
Founders Tilman Flock (CEO), Martin Copes, and Zaki Trache were core contributors to Palantir's AIP and spent years inside its life-sciences practice before leaving to build Perceptic with a team of roughly 20. Flock describes legacy drug development as a chain of handoffs: "That's a linear process where insight dies at every handoff." Perceptic positions itself as the connective tissue between discrete AI tools and the proprietary internal and external data that pharma companies use to make go/no-go calls.
What ships
The platform breaks into three modules:
- Scout — external asset triage for licensing candidates and competitor programs. The company says it cuts scientific due diligence from roughly a week to an hour and scales screening from hundreds of assets per week to thousands in minutes.
- PercepticOS — an internal intelligence layer for hypothesis testing and evidence comparison across a company's own data.
- Atlas — a clinical-data foundation for trial design that, in production, reportedly delivered a 50-fold increase in clinical-data extraction throughput.
Traction
Perceptic says it is already deployed with multiple top-tier pharmaceutical companies, biotechs, and contract research organizations, though the only customer it is permitted to name is Australian biotech CSL.
Why it matters for builders
The signal here isn't the dollar figure — $12 million is modest in a year of billion-dollar rounds. It's the architecture bet: that the bottleneck in enterprise AI for regulated industries is integration, data context, and governance, not raw model capability. Ex-AIP operators are wagering that the durable layer sits between foundation models and proprietary enterprise data — the same wedge Palantir drove in defense. Treat the LLM as swappable; own the data plumbing and the audit trail.
That puts Perceptic in a different lane from model-first drug-discovery plays like Isomorphic Labs or Profluent. It isn't training a foundation model for biology; it's selling orchestration software that makes whatever model you license usable against regulated, fragmented pharma data. For anyone shipping vertical AI into pharma, finance, or legal, the template is the same: capture the workflow and the provenance layer, and the model underneath becomes a commodity input.
