All cases

Case 02 — 2026

Legal Copilot

RAG over contracts

Reviewing an incoming contract took a lawyer up to three hours. The copilot checks the document against company templates and policy and returns a list of deviations — each one linked to a specific clause.

Client
Property developer, 400+ construction contracts a year
Industry
Construction & real estate
Timeline
8 weeks, including a pilot on 50 contracts
Team
2 engineers, an in-house lawyer as domain expert

Outcome

3 h → 15 min
for a first pass on a contract
94%
of deviations found automatically
100%
of conclusions linked to a clause

The task

A team of four lawyers reviewed contracts by hand against a 60-point checklist. Approvals stretched across a week, and the risk of missing a clause grew towards the end of the day.

  • Up to 3 hours for a first pass on one contract
  • A 60-point checklist, all of it verified by eye
  • Templates, policy and past opinions lived in three separate folders
  • “The model said so” is not an answer a lawyer accepts

What we built

The hard requirement was not accuracy but verifiability. So the system never paraphrases the document: every conclusion comes with a quote and a location in the file.

  • A base of 1,200 documents: templates, policy and past opinions
  • Parsing of PDF, DOCX and scans, including tables and annexes
  • A separate check for each of the 60 checklist points
  • Deviations with a quote, a page number and a risk level
  • Access control: a lawyer only sees documents from their own projects

How it works

  1. 01

    Contract upload

    PDF, DOCX or a scan enters the system; scans go through recognition first.

  2. 02

    Structure mapping

    The document is split into sections and clauses so every link points to an exact spot.

  3. 03

    Check against the base

    Each checklist point triggers a search across company templates and policy.

  4. 04

    The opinion

    A list of deviations by risk level, with quotes and suggested wording.

Stack

  • Anthropic
  • LlamaIndex
  • Qdrant
  • pgvector
  • OCR
  • FastAPI

“Our lawyers trust the answers because every one of them links to a specific clause in a document.”

Head of Legal

What's next

Next up is the counter-proposal: the system drafts the amended wording and the lawyer approves or rewrites it.

A similar task?

We'll build a prototype for your process in one day. Free, with no strings attached.

Next caseRequest Intake