Liveproduct
Clausebook
HIPAA redlining playbooks the lawyer owns.
Build a Business Associate Agreement redlining playbook through an adaptive interview, then apply it to a real uploaded BAA. Clausebook splits the document into clauses, classifies each against an ontology and your playbook, and drafts redlines cited to HIPAA authority.
- HIPAA BAA redlining
- Playbook you own
- Cited to eCFR
- Live now
- Ontologies & knowledge graphs
- NLP & information extraction
- Regulated & high-security environments
- Data science & evaluation
- typed PageRank concept-importance correlation (ρ)
- 0.830
- held-out false-redline rate, down from 0.607
- 0.018
- test files, approximately
- 118
The problem
BAA negotiation is repetitive expert judgment trapped in one person’s head. Outsourcing it loses the positions you fought for. Keeping it in-house caps it at whoever holds that knowledge.
The approach
Clausebook is built for ownership rather than automation. The playbook is the lawyer’s asset and the tool applies it. Every accepted edit writes back, so using the product is how the playbook improves. Clausebook is a software tool, not a law firm, and does not give legal advice.
How Clausebook fits together
Interview
Compiles your playbook
Upload
A real BAA, extracted server-side
Classify
Clauses against the Atlas ontology and your playbook
Redline
Cited to eCFR text and case law
Accept or edit (a person decides)
The lawyer decides
Every accepted edit writes back to the playbook
What Clausebook does
The playbook is yours
Every accepted edit writes back to the playbook you own, so using the product is how it improves.
Redlines cited to authority
Drafted with live citations from eCFR regulation text and CourtListener case law.
Criticality over frequency
A typed legal-dependency graph ranks concepts by importance, not by how often they appear.
An aggression guard that holds
Cuts held-out false redlines from 0.607 to 0.018 at full required recall.
What’s inside Clausebook
- Ingest by paste or PDF, Word or text upload, extracted server-side
- Clause splitting, then classification against the Atlas ontology plus the tenant’s own playbook
- Redline drafting grounded in live citations from eCFR regulation text and CourtListener case law
- Accept or edit, with each edit writing back to the playbook: the ownership loop
- Corpus mining that surfaces which of your past positions are well-evidenced and which are inconsistent
- Word export with native tracked changes
- Per-tenant isolation via row-level security
Where Clausebook stands today
- Roughly 118 test files, with CI, auth, row-level security and billing in place
- A typed legal-dependency graph that ranks concepts by importance rather than frequency. Typed PageRank recovers clause criticality at ρ = 0.830, against 0.639 for prevalence alone
- An over-aggression guard that cuts held-out false redlines from 0.607 to 0.018 at full required recall
The Clausebook tech stack
- React 18
- Vite
- Python
- FastAPI
- serverless
- Claude
- Supabase Postgres
- Supabase Auth
- row-level security
- Stripe
Try Clausebook
Live now, with credit-based pricing.
Start building a playbook (opens in a new tab)