01
The firm thesis
Legal track
Who you serve, what you sell, and where judgement sits.
Technical track
Map the system boundary: what the platform does and does not do.
OutputOne-page firm thesis
The flagship · 10 days
Build Your Own AI-Native Law Firm.
Ten applied days spanning business, practice and platform. Lawyers and technical collaborators work on parallel tracks around one fictional family-law matter, and finish with a working firm design and a reference system they own.
Format, dates and fee: in development — register interest
We are considering a two-week intensive and a part-time format. Neither is confirmed.
Legal track
For lawyers and firm leaders.
Service design, vocabulary, review standards, supervision and the firm's operating model. No coding — but you will read and judge what the system does.
Technical track
For builders working beside lawyers.
Lovable, Supabase, SurrealDB, Voyage AI and OpenAI/Claude via Amazon Bedrock — configured, connected and evaluated.
Day by day
02
Service and application design
Legal track
Design the client journey for one service line.
Technical track
Distinguish original application work from reusable components and purchased services.
Build, adapt or integrate
Build, adapt or integrate: decide for each capability.
OutputArchitecture and ownership decision
03
Vocabulary and taxonomy
Legal track
Write down how your firm names matters, parties and work.
Technical track
Compare a small noslegal subset with the firm's terms.
Build, adapt or integrate
Adapt shared concepts; accept, modify or exclude each.
OutputConcept mapping
04
Ontology and SurrealDB
Legal track
Define the questions the firm must be able to answer.
Technical track
Model concepts and relations explicitly in SurrealDB.
Build, adapt or integrate
Build the matter graph from the mapping.
OutputCanonical identifiers, relations and test questions
05
Documents and retrieval
Legal track
Decide which document facts matter and how they are checked.
Technical track
One prepared extraction pipeline with Voyage AI retrieval and source references.
Build, adapt or integrate
Integrate extraction; keep every citation.
OutputReviewed extracted records and corrections
06
The matter workflow
Legal track
Supervision, review points and client communication.
Technical track
Interface and data layer in Lovable and Supabase.
Build, adapt or integrate
Build the first working screen around one matter.
OutputA working matter view
07
Agent boundaries
Legal track
What an agent may never do without a lawyer.
Technical track
One read-only connector with explicit authorisation and tool contract.
Build, adapt or integrate
Integrate a connector — and prove a denied call fails safely.
OutputTool contract and a recorded denied test
08
Lab build
Legal track
Test the workflow against the matter pack.
Technical track
Integrate a single selected component where it serves the use case.
Build, adapt or integrate
Build or adapt the capstone feature.
OutputWorking adapter and recorded result
09
Evaluation
Legal track
Review outputs as a supervising lawyer would.
Technical track
Compare against an explicit baseline and answer set.
Build, adapt or integrate
Decide: retain, adapt or reject each component.
OutputResults, errors and decisions
10
Operating the firm
Legal track
Regulatory posture, insurance questions and the next ninety days.
Technical track
Handover, ownership of accounts, and maintenance.
OutputPresented firm plan and next actions
The matter pack
One fictional family-law matter, carried through all ten days.
Entirely invented. No client data is used at any point. The matter is deliberately imperfect, because real matters are.
- —Client instruction note and initial attendance record
- —A short chronology with deliberate gaps and one contradiction
- —Financial disclosure extracts and correspondence
- —A supplied source pack for the relevant law of England and Wales
- —A supervising lawyer's answer set, used for evaluation
How work is assessed
Judged as a supervising lawyer would judge it.
- 1Does the system answer the matter's test questions — with sources?
- 2Are errors found, recorded and corrected, not hidden?
- 3Is every agent action bounded, and does a denied call fail safely?
- 4Could a supervising lawyer explain how each output was produced?
- 5Does the participant own the accounts, data and code?
What you leave with
- A one-page firm thesis and service design
- A firm vocabulary mapped to shared legal concepts
- A matter ontology with canonical identifiers and test questions
- A working reference application on your own accounts
- An evaluation record: baseline, answer set, errors, decisions
- An operating plan covering supervision, data handling and review
Register interest in the flagship.
Free, no commitment. We write when dates and terms are confirmed — and not otherwise, unless you ask us to.
Register interest →