Claude integrations
Anthropic models applied where long-context reasoning and reliability matter most.
- Engagement
- Fixed-scope engineering on a defined document set
- Typical timeline
- 5–7 weeks
- Starts with
- Three documents your team has already reviewed, with their conclusions, so there is a known answer to measure against.
A 340-page facility agreement arrives, and someone has to find every clause touching early repayment, then check each one against a policy document that runs to sixty pages of its own. Nothing about the task is creative; it is reading, at a length that defeats human attention by mid-afternoon.
Dependable outputs on the tasks that carry real risk.
What we build
Concrete artefacts, handed over and documented.
Long-document handling that keeps the whole agreement in view instead of cutting it into fragments and hoping the right one surfaces
Caching of the policy text that repeats on every run, which is usually where the cost of this work sits
Findings that quote the clause and give the page, so a lawyer can verify in seconds rather than re-reading
A strict output shape, asserted on every response, if the next step expects six fields in a fixed order, that is tested, not trusted
A refusal path for documents outside the agreed scope, so an unusual contract is flagged instead of forced into the template
What changes
Review of long documents drops from days to an hour of checking
Every finding traces back to a page, which is what makes the output usable in a legal or audit setting
Repeat runs against the same policy set cost a fraction of the first
How it runs
- Week 1
01Worked examples
We take three documents your team has already reviewed and turn their conclusions into the standard we measure against.
- Weeks 2–4
02Build and cache
Document handling, citation extraction and the caching that makes repeated runs affordable.
- Weeks 5–7
03Verify
Run against the reviewed set, count what it missed rather than what it found, and tighten until the misses are acceptable.
Chosen per project. Named here so you can see the shape of it.
- Anthropic Claude
- Vercel AI SDK
- Postgres
- Langfuse
Questions we get asked
Tell us the outcome, not the tooling.
Send us the situation you are in. We will tell you which discipline it belongs to, what we would do first and what it costs, including when the answer is to wait.