From documents to defensible findings. Sources traceable, doubts flagged.
A research commercialisation project. Funded by
Underwriting, diligence, compliance, research.. So much of it comes down to finding the right facts in messy, scattered sources. Chat-based AI tools offer help, but they work like black boxes: confident answers with no honest signal of what they got wrong.
Loses the thread across long document sets. Citations inconsistent, often hallucinated. Confident output regardless of underlying certainty.
Reads documents exhaustively and works across many documents. Backs claims with auditable source citations. Model abstains and asks for clarifications when uncertain or there is contradictory information.
Drop in the source documents: filings, reports, transcripts, spreadsheets. Any format, any scale.
Spell out what the reports or structured outputs should contain: the fields to extract, the shape of the output, the constraints it must satisfy.
Get the report or structured output back with a full audit bundle: every field linked to the source span it came from.
We pair generative LLMs with discriminative zero-shot models that identify the relevant chunks and spans before generation. This anchors every answer to a specific location in the source and produces calibrated confidence, something a single LLM cannot reliably do on extraction tasks.
Workflows operate over whole collections, not single files. When several passages are relevant or sources contradict one another, the system reconciles them explicitly and surfaces the conflict, rather than silently picking one and moving on.
The interface is the proof. Every answer traces back to a highlighted span in its source document, with reconciliation steps and conflicting information shown in full, turning verification from a re-investigation into a glance.
ReliaParse is created by a team of academics and commercial experts based at the University of Helsinki.
15+ years of research in computational social science, including Oxford and University College London. Leading research work.
Widely published in statistical methods and LLM applications. Industry experience as a data scientist.
Runs pilot relationships, scopes engagements with partner organisations, and maps the route out of the research phase.
20+ years of entrepreneur in tech related startups & corporate venturing. Coordinating go-to-market & AI risk mitigation.
Background in basic research in signal processing, and expertise in commercialisation in the digital domain.
Not yet. ReliaParse is a research project actively developing new approaches to document AI, and we’re working with a small number of design partners to evaluate the system on real, high-stakes work. Early users get hands-on access to the current system and direct involvement in shaping where it goes next. We expect a productised offering to follow, but design partners come first.
PDFs, scans, Word documents, spreadsheets, and emails: anything that arrives in a typical case file. We handle native digital documents and scanned ones, and we work with documents in European languages such as Finnish. If your workflow involves a format we don’t yet support, tell us, and we will include it in our roadmap.
The pilot version of our system runs on CSC infrastructure in a data centre in Kajaani, Finland, not on the public cloud. Data shared during a pilot is not used for training models nor any other purposes. We work within whatever data handling arrangement potential users require, and are open to on-premise deployment for sensitive workflows where needed. We’re happy to sign a DPA before any documents are shared.
ReliaParse is research project based at the University of Helsinki. We are funded by Business Finland as a Research to Business project. Our team’s background is in applied machine learning in the field of computational social sciences, especially NLP and the reliable processing of heterogenous data.
Could your work benefit from new approaches to AI? Get in touch to discuss how we would collaborate. We are looking for early users to help us develop the project following user needs. We are able to offer our system free for early users.