sci2sci Raises €1.2M for Auditable AI

sci2sci funding has brought €1.2 million in pre-seed capital to a Berlin startup building a traceability layer for artificial intelligence in regulated industries. The round was co-led by Heliad and IBB Ventures, with Robin Capital and Superangels participating. The useful question is not whether the system can produce more text, but whether a customer can trace each important claim back to admissible evidence.

Key takeaways

sci2sci announced a €1.2 million pre-seed round on September 8, 2026.
The company says it will expand engineering and deployments of VectorCat and Integrity Cortex, beginning with biopharma.
Funding validates investor interest, not regulatory compliance or product accuracy; buyers still need independent validation.

sci2sci funding: the verified transaction

IBB Ventures published the direct announcement dated September 8. Heliad separately explained its investment rationale. Tech.eu, Dealroom News, StartupCentrum and Bolus recorded the same round, amount, leads and participants. These sources support the transaction; they do not disclose valuation, ownership, revenue, cash burn or the exact commercial terms.

Verified sci2sci funding facts
Round	€1.2 million pre-seed
Co-leads	Heliad and IBB Ventures
Other participants	Robin Capital and Superangels
Company	Berlin-based sci2sci
Products named	VectorCat and Integrity Cortex
Initial sector	Biopharma, with other regulated sectors planned

The distinction between disclosed facts and company claims matters. The funding amount and investor list are supported across the current-event sources. Statements about product safety, customer deployments, market leadership and future expansion originate with the company or its investors and remain attributed. No source provides an audited customer count or independently measured error rate.

What sci2sci says it is building

sci2sci describes VectorCat as a data-integration layer that finds information across cloud storage, network drives and laboratory systems without forcing an immediate migration. Integrity Cortex is presented as the control layer that links conclusions to sources and records the reasoning path. The company says both products use Parseltongue, a framework it released under the Apache 2.0 licence.

That architecture addresses a real organisational problem. A regulated company may store study reports in PDFs, measurements in spreadsheets, protocols in laboratory systems and approval records in another repository. A generic language model can retrieve fragments, but retrieval alone does not prove that the right version, context or authority was used. A credible system must preserve provenance, version history, permissions and the boundary between evidence and inference.

The company’s answer is a structured network in which claims remain connected to their sources. If implemented as described, that can make review faster because an auditor or scientist does not have to reconstruct every answer from scratch. Yet the value depends on connector coverage, identity controls, source quality and how the system behaves when evidence conflicts. A neat citation is not enough if it points to an obsolete procedure.

Why biopharma is the first test

Biopharma is a demanding entry market because decisions can affect patients, trials, manufacturing and regulatory submissions. Records may need to remain available for years, and a company must demonstrate who changed data, when it changed and why. IBB Ventures says sci2sci is already working across preclinical research, contract clinical research and bioprocess operations, but names and measured outcomes are not disclosed.

The reference to 21 CFR Part 11 in the announcement should be read carefully. Part 11 establishes requirements around trustworthy electronic records and signatures in the United States. A software vendor can design features to support compliance, but software does not make a customer compliant by itself. Validation, procedures, training, access control, audit review and the customer’s intended use remain part of the compliance system.

This is the mechanism behind the investment angle: companies want AI assistance without losing the evidence trail required for high-consequence decisions. The commercial test is whether sci2sci can reduce review effort while preserving, or improving, the quality of decisions. Buyers should demand a controlled before-and-after study, including exception rates and the time humans spend resolving unsupported or contradictory results.

What the capital can reasonably change

The announcement says the money will expand engineering, deepen deployments and reach more biopharma customers before extending into sectors such as banking. At pre-seed stage, those goals are plausible uses of capital, but they are plans rather than completed results. Hiring can improve connector coverage and implementation support; it can also increase fixed costs before recurring revenue is proven.

Deployment depth matters more than the number of logos attached to a slide. A regulated customer may begin with a narrow document set and a small user group. Moving into production requires security review, data-processing agreements, validation evidence, change-control procedures and support for incidents. Each step can lengthen sales cycles. The startup needs enough runway to learn from these controls without promising that every pilot will become a large contract.

Expansion into banking would introduce different vocabularies, authorities and retention rules. The underlying traceability problem is similar, but a connector or ontology designed for laboratory records cannot simply be relabelled for lending or compliance. The company will need domain-specific review, testing and governance. Its open-source foundation may help technical inspection, although customers must still evaluate the proprietary layers that run in production.

The questions customers should ask

First, buyers should ask what the system means by a verified claim. Does verification confirm that a citation exists, that the cited passage entails the answer, that the source is current, or all three? These are different checks. A model can cite an authentic record and still misstate what it says. Evaluation should therefore include contradictory evidence, missing documents and deliberately misleading inputs.

Second, customers should ask how permissions propagate. An assistant that can search many repositories must not reveal a restricted document merely because it can infer the answer from related data. Access controls need to follow the user, source and purpose of the request. Logs should allow an investigation to reconstruct which records were available at the moment a response was produced.

Third, the organisation needs a response for uncertainty. A safe system should abstain or escalate when evidence is insufficient. That behaviour can feel slower than a confident chatbot, but it is central to regulated work. Commercial claims should disclose how often the tool abstains, how frequently reviewers overturn its output and which categories of work remain outside the supported scope.

Finally, buyers should separate technical evidence from business evidence. A successful benchmark can show that a model retrieves citations correctly under test conditions. It does not establish integration cost, user adoption, renewal or savings in a live process. The next useful disclosure would describe a bounded production workflow with a baseline, review population, error definition and measured outcome.

Why the sci2sci funding story matters

Everyone else is reporting a €1.2 million pre-seed round; we are explaining the control system that must sit between a language model and a regulated decision. The pitch is strongest where a company can show that evidence remains connected, current and inspectable throughout the workflow. It is weakest when “trustworthy AI” becomes a broad label without measurable limits.

Investors are backing a category built around verification rather than raw model capability. That could be durable because regulated customers cannot simply trade oversight for speed. It could also be difficult because every deployment exposes differences between repositories, teams and rules. A small startup must avoid becoming a custom-services business while still doing enough integration work to prove value.

The open-source Parseltongue layer gives developers something concrete to inspect, but the commercial proposition rests on operating controls around it. Customers should examine licensing, maintenance responsibilities, security updates and the boundary between open and proprietary components. They should also test whether exported evidence records remain usable if they later change vendors.

Readers can compare this evidence-first approach with Lapaas Voice coverage of Blee’s compliance-focused AI funding and Hope Care’s regulated remote-monitoring expansion. In each case, capital supports a system that touches sensitive decisions, so credible growth depends on boundaries, auditability and measurable performance rather than a funding headline alone.

The current record supports a fresh financing event and a clearly defined product thesis. It does not support a valuation, revenue estimate or claim that regulators have approved the platform. Those gaps are important. A future update should be triggered by a distinct event such as a disclosed production deployment, independently measured result, certification or new financing—not by another article repeating the same announcement.

Procurement teams should also define portability before deployment. Evidence links, audit logs and review decisions may become business records that must survive a vendor change. A useful contract should specify export formats, retention, deletion, incident notice and the treatment of model or connector updates. Those operational details determine whether traceability remains available when the underlying software changes.

FAQ

How much did sci2sci raise?

sci2sci raised €1.2 million in a pre-seed round announced on September 8, 2026.

Who invested in sci2sci?

Heliad and IBB Ventures co-led the round. Robin Capital and Superangels also participated.

What does sci2sci build?

The company builds software intended to connect fragmented enterprise data and keep AI-generated claims linked to evidence, initially for biopharma and other regulated work.

Does the product guarantee regulatory compliance?

No. Software can support record controls, but compliance also depends on validation, procedures, access management, training, review and the customer’s intended use.