sci2sci, a Berlin startup founded by Angelina Lesnikova and Valerii Kremnev, has raised €1.2 million in pre-seed funding co-led by Heliad and IBB Ventures, with Robin Capital and Superangels also participating, according to Tech.eu’s September 8, 2026 report. The company’s Integrity Cortex product links AI outputs and company documents into a network where every claim must be traced to a source, targeting the audit-trail requirements of 21 CFR Part 11 in biopharma before expanding into banking.
€1.2 million is a rounding error next to what actual Part 11 compliance work costs. A parallel item from PRWeb, dated August 26, 2026, shows REJIMUS expanding manual Part 11 validation services — gap assessments, IQ/OQ/PQ test-protocol execution, supplier audits — for consumer-product companies that don’t even touch AI yet. That’s the existing cost structure sci2sci is implicitly selling against: FDA auditors still want executed test protocols and human sign-off, and it’s unclear from Tech.eu’s reporting whether an auditable claim-graph generated by software satisfies that bar or simply produces better documentation to hand a validator like REJIMUS.
The traceability fight is bigger than one startup
A companion piece from Quality Digest (dated June 10, 2026) captures the live industry argument sci2sci is stepping into: AI coding assistants reportedly generate 40% or more of new code in enterprise environments, and critics in that piece argue regulated software shouldn’t use AI-generated code at all if nobody can fully explain why it failed. sci2sci’s pitch — that its systems have no primitive for an “ungrounded statement” — is an architectural answer to that exact objection, not a regulatory one.
sci2sci’s real product isn’t the software — it’s a bet that regulators will eventually treat machine-generated provenance as equivalent to human-executed validation, and that bet is still untested.
Watch for two things: whether sci2sci names actual biopharma customers and revenue beyond the vague “preclinical research, contract clinical research and bioprocess operations” description Tech.eu reported, and whether the banking expansion produces a comparable evidence-per-claim standard in a sector with a very different regulator and a much lower tolerance for open-source dependencies like the Apache 2.0-licensed Parseltongue framework underneath it.
Instead of shipping capabilities first and patching trust and security afterward, we built Parseltongue and Integrity Cortex to permit only safe outputs. A model’s lockpicking skills are useless when there’s no door — and our software guarantees there is none.
— Tech.eu
