AI Legal Research Portal — 70% Faster Case Prep for a European Law Firm
Legal
How Visual-AI-Labs delivered a private AI research portal over case law, internal precedents and contracts in two 30-day cycles, cutting case-prep time by ~70%.
- −70% — Case-prep time
- < 3 sec — First useful answer
- 100% (no answer without citations) — Sources cited
- 92% of fee earners — Adoption (8 weeks)
The problem
The firm’s lawyers spent 6–10 hours per matter on initial research: reading judgments, hunting through internal memos and old contracts on shared drives, and re-deriving arguments already used in prior cases. Knowledge was siloed by partner and by year; juniors re-discovered the same precedents repeatedly. Off-the-shelf legal AI tools were either US-centric or required uploading client documents to non-EU clouds, which the firm’s general counsel had ruled out on GDPR grounds.
What Visual-AI-Labs built
Visual-AI-Labs delivered a private AI research portal in two 30-day cycles. Cycle 1 shipped the core: secure ingestion of the firm’s case archive, contracts and internal memos into a vector store, a retrieval-augmented chat interface with mandatory inline citations, and SSO against the firm’s identity provider. Cycle 2 added matter-scoping (each conversation is bound to a matter and its allowed sources), a "brief generator" that drafts a structured argument outline from a question, and an admin console for the knowledge officer to manage sources, redactions and retention.
- Private RAG over case law, internal memos and contracts — EU-hosted
- Inline citations on every answer; no answer is returned without sources
- Matter-scoping and per-user access controls (mirrors the firm’s conflict rules)
- Drafting assistant for structured argument outlines and clause comparisons
- Admin console: source management, redactions, retention, audit log
Results
Eight weeks after launch, fee earners reported a ~70% reduction in time spent on initial case research. Partners noted that junior lawyers now arrive at the first internal review with a structured outline and cited sources, rather than a blank page. Adoption reached 92% of fee earners without a mandate — the portal simply became the fastest way to start a matter.
Process & timeline
Visual-AI-Labs ran two successive 30-day cycles with a weekly demo. The first cycle prioritised a working end-to-end loop on a single practice area; the second generalised it across the firm and added the drafting assistant. No discovery phase ran in isolation — discovery, build and feedback happened inside the same cycles, which is how Visual-AI-Labs keeps timelines short without sacrificing quality.
Discuss a similar project with Visual-AI-Labs →
FAQ
Where is the data hosted?
Inside the EU, on infrastructure controlled by the firm. Visual-AI-Labs does not subcontract development or hosting outside the EU.
Which AI models were used?
A combination of EU-available frontier models for reasoning and an open-weights embedding model for retrieval. The portal is model-agnostic — the firm can swap models without a re-architecture.
How are hallucinations handled?
The portal does not return an answer without inline citations to its own indexed sources. If retrieval returns nothing relevant, the portal says so instead of guessing.
How long did the project take?
Two successive 30-day cycles — about 9 weeks from kickoff to firm-wide rollout, including the second cycle’s drafting assistant.
Was the firm’s case archive a problem?
No. Ingestion (PDFs, DOCX, scanned judgments with OCR) was handled in the first cycle. Visual-AI-Labs treats messy archives as the normal starting point.
Can it be extended to other practice areas?
Yes — the architecture is matter- and practice-scoped from day one. Adding a new practice area is a configuration change, not a rebuild.
Did it integrate with the firm’s DMS?
Yes, via a thin connector built in cycle 2. SSO uses the firm’s existing identity provider.
How can a similar project start?
Through a short scoping call with Visual-AI-Labs. The first 30-day cycle typically begins within 2–3 weeks of that call.