A Berlin software company that wants to make artificial intelligence auditable in highly regulated industries has secured €1.2 million in pre-seed capital. The round, announced on 8 September, was co-led by Heliad and IBB Ventures with participation from Robin Capital and Superangels. For a European deep-tech startup, the sum is modest, but the problem it addresses, how to trust AI outputs when regulators demand a complete evidence trail, is anything but.

The regulatory pressure behind the funding

European regulators are finalising rules that will require providers of high-risk AI systems to demonstrate transparency, traceability and human oversight. The EU AI Act classifies AI used in pharmaceutical research, clinical evaluation and credit scoring as high-risk, meaning developers must prove their models do not hallucinate and that every conclusion can be traced to a verified source. At the same time, the European Medicines Agency expects electronic records in drug development to comply with standards equivalent to the US 21 CFR Part 11 framework. sci2sci positions its Integrity Cortex product directly at this intersection.

The startup was founded by Angelina Lesnikova and Valerii Kremnev, who previously worked on data integrity problems in life-sciences R&D. They observed that research data lives in PDFs, spreadsheets, laboratory information management systems and informal notes, none of which talk to each other. When an AI model is layered on top, the risk of ungrounded assertions multiplies. Their response was to build a verification layer first, then add AI capabilities on top of it.

How Integrity Cortex and Parseltongue work

Integrity Cortex connects documents, data sets and AI-generated outputs into a structured knowledge graph. Every claim, whether written by a scientist or produced by a large language model, must be linked to a source that the system can verify. The underlying framework, Parseltongue, was open-sourced this year under the Apache 2.0 licence, a move intended to let external developers audit the verification logic and build connectors for proprietary laboratory systems. A second product, VectorCat, indexes files across cloud storage, network drives and lab instruments without moving the underlying data, creating a searchable catalogue that both humans and AI agents can query.

Kremnev describes the architecture as a "door" that simply does not exist for ungrounded statements. The model is forced to produce evidence for every claim; if it cannot, the output is rejected. That approach contrasts with the prevailing industry pattern of deploying models quickly and adding guardrails later. Whether the performance penalty is acceptable for day-to-day research workflows remains an open question the company will need to answer as deployments scale.

Early traction in biopharma

The company says it is already working with customers in preclinical research, contract clinical research organisations and bioprocess operations. These are environments where a single missing citation can delay a regulatory submission by months. By embedding verification into the data layer, sci2sci aims to reduce the manual effort of assembling audit packages. The funding will be used to grow the engineering team, deepen those existing deployments and sign additional biopharma accounts.

Expanding into financial services

Beyond life sciences, sci2sci plans to adapt Integrity Cortex for banking regulation, where model risk management rules such as the ECB's guide to internal models and the Basel Committee's principles for effective risk data aggregation demand similar traceability. The technical requirements overlap: immutable audit logs, version-controlled data lineage and the ability to explain automated decisions. If the platform can serve both sectors with a single codebase, the addressable market expands considerably.

Competitive landscape and open questions

Established electronic lab notebook vendors such as Benchling and Dotmatics are adding AI features, while enterprise search platforms like Glean and Sinequa market retrieval-augmented generation for regulated industries. sci2sci's bet is that verification cannot be bolted on, it must be architectural. The open-source release of Parseltongue is a strategic lever: if the framework becomes a standard for evidence graphs, the company benefits from network effects. But adoption by conservative IT departments in pharma and banking is rarely fast, and the €1.2 million runway will need to stretch through long sales cycles.

What happens next

People mentioned

  • Valerii Kremnev

    Co-founder and CTO, sci2sci

  • Angelina Lesnikova

    Co-founder, sci2sci

Organisations

sci2sci · Heliad · IBB Ventures · Robin Capital · Superangels