Sci2sci Raises 1.2 Million Euros for Trustworthy AI in the Biopharma Industry

The Berlin-based AI startup Sci2sci has closed a pre-seed funding round of 1.2 million euros. The round is co-led by Heliad and IBB Ventures, with Robin Capital and Superangels also participating. The company is developing a neurosymbolic infrastructure designed to make the use of AI in highly regulated industries more transparent and verifiable. Sci2sci is initially focusing on the biopharmaceutical industry.

With the fresh capital, the startup plans to expand its engineering team and accelerate the rollout of its two products, VectorCat and Integrity Cortex. In the future, the technology is expected to be used not only in life sciences but also in other regulated industries, such as the banking sector.

Heliad and IBB Ventures Lead Funding Round

The pre-seed round is led by Heliad and IBB Ventures. Robin Capital and Superangels are also participating.

Most companies selling AI for regulated industries are banking on language models becoming accurate enough.

Christopher Garlich, Investment Lead at Heliad

Sci2sci takes the approach of preventing unsubstantiated statements right at the architectural level.

Sci2sci thus addresses a problem that is becoming increasingly significant with the growing use of generative AI in enterprises: Language models can process large datasets, but their results are of limited use for regulated processes if statements and conclusions cannot be reliably traced back to their original source.

Biopharma as the First Target Market

This issue is particularly relevant in the pharmaceutical industry. Research results, lab values, clinical data, and regulatory documentation are often scattered across different systems, PDFs, spreadsheets, or lab records. For example, if statements within a regulatory submission need to be verified, tracing a piece of information across multiple documents can lead all the way back to the underlying raw data.

AI is not intended merely to summarize documents, but to link statements to their sources and formally verify conclusions. The company will also be supported in this effort by Stephanie Bova, former Corporate Vice President and Digital Transformation Officer for R&D at Novo Nordisk. She has joined Sci2sci as a Senior Advisor.

“Knowledge as Code” Instead of Traditional AI Summarization

At the heart of the technology is Integrity Cortex. The product follows the “Knowledge as Code” principle.

To this end, information from documents, corporate data, and AI outputs is organized into an interconnected knowledge network. Instead of simply having a language model generate continuous text, each statement is linked to a verbatim citation from the underlying source.

Conclusions, in turn, must be derivable from previously defined statements. A symbolic component then verifies sources and derivations; if the AI model invents a piece of information, the verification should fail. At the same time, the system can flag which additional conclusions depend on the erroneous information. According to the company, if an input document changes, it is also possible to trace which downstream statements are affected.

Neuroscience Meets AI

This approach is also tied to the scientific background of co-founder and CEO Angelina Lesnikova. She earned her Ph.D. in neuroscience and studied the molecular mechanisms of memory and learning. This led to the basic idea of treating corporate knowledge not as a static data repository, but as an active network of interconnected information.

CTO and co-founder Valerii Kremnev is responsible for the technological implementation of this approach. A key foundation is the Parseltongue framework, which Sci2sci released as open source this year under the Apache 2.0 license. In August, a team of researchers using a system based on this framework took second place at a biopharma AI hackathon. The system evaluated potential drug targets against scientific literature and clinical studies.

Auditability Is Becoming a Decisive Factor

For regulated industries, a plausible AI response is not enough. Companies often need to be able to document how information was generated and on what data decisions are based. Sci2sci therefore describes its verification layer as deterministic and interpretable. At the same time, the goal is to create a complete audit trail that is compatible with the requirements of 21 CFR Part 11 for electronic records.

The company is thus targeting, among other things, regulatory submissions, production documentation, and other processes in which statements must be unambiguously traced back to their source. This approach thus differs from traditional retrieval or enterprise chatbot systems: the focus is not solely on finding the right information, but on formally verifying how a statement came to be.

VectorCat connects scattered enterprise data

Integrity Cortex is complemented by VectorCat. The data integration layer connects cloud storage, network drives, and laboratory systems without requiring companies to first fully migrate their existing data. VectorCat is designed to create a searchable catalog from this data for employees and AI agents. Integrity Cortex then takes on the task of making the information derived from this catalog traceable and verifiable. According to Sci2sci, the products are already in productive use by customers across various areas of drug development, ranging from preclinical research to contract clinical research and bioprocessing.

Sci2sci plans to use the 1.2 million euros initially to expand existing implementations and acquire additional customers in the biopharmaceutical industry. At the same time, Integrity Cortex will be further developed for other regulated industries. The company cites the banking sector as a potential next market. There, too, information often needs to be traced across large volumes of interconnected documents and documented for regulatory purposes.

The Next AI Competition Could Be About Trust

Sci2sci’s funding round reflects a shift in the enterprise AI market. After companies have tested numerous generative AI applications in recent years, questions of traceability, governance, and actual production readiness are increasingly taking center stage. Particularly in the pharmaceutical, financial, and other regulated industries, it is not enough for an AI system to be correct most of the time. What matters is that companies can trace and document why a statement is correct and from which source it originates.

Sci2sci does not attempt to solve this problem by adding an additional layer of control at the end of the process, but rather through the architecture of the system itself. It remains to be seen whether this neurosymbolic approach will prevail over other forms of retrieval, knowledge graphs, and AI governance systems. With its first production-ready customers and pre-seed funding, Sci2sci now faces the key question of whether its technology can scale beyond individual biopharma applications.