cashcrown // case study · legaltech / produkt własny
A local system that turns raw case files into structured evidence where every claim traces back to a file, SHA256, page and quote — because the LLM is not the source of truth.
Case files are a mess of formats — PDFs, scans, DOCX, e-mails — and the work means manually digging through them to find contradictions and establish who knew what and when. Language models are tempting but hallucinate: they assert a 'fact' that isn't in the files, and there's no way to trace it back to a source. There was no tool that combined automation with the hard rule that every claim must have evidence.
Project timeline
Select a node to see its description and data flow.
An advanced, hardened prototype (v0.1.0, MIT) — phases A-E, 7 hardening cycles, ~873 tests and CI, ~14.8k source spans on prod data. It's an analytical tool for professionals (CLI/venv), not legal advice or a replacement for a lawyer: without an LLM only the rule-based mode runs, OCR is optional, and ~72.7% precision needs a human in the loop. Phase F is under way: export to legal formats and scale.