Product discovery · legal workflow · AI-assisted build
Matter
The useful part of this project is not that I built a legal-document workbench. It is that the first product idea was wrong, the attorney said so, and the build changed.
The premise died in discovery
Built an attorney-intake concept.
The premise was that intake was the painful workflow worth solving.
The premise was that intake was the painful workflow worth solving.
A practicing family-law attorney was not meaningfully interested.
That was the useful result. I dropped attachment to the first concept instead of defending it.
That was the useful result. I dropped attachment to the first concept instead of defending it.
Asked what she actually needed.
The real job was discovery/document work: making a messy packet inspectable without allowing polished synthesis to outrun the source.
The real job was discovery/document work: making a messy packet inspectable without allowing polished synthesis to outrun the source.
Rebuilt substantially in roughly 1–2 weeks.
The product became a working legal-document workbench rather than an intake tool with better positioning.
The product became a working legal-document workbench rather than an intake tool with better positioning.
What the workbench does
- Python document pipeline with PDF ingestion and OCR fallback.
- Source-quality grading and conflict detection across filings and orders.
- Consequential fields linked back to source evidence and provenance.
- Deadlines remain blocked when case posture is unresolved rather than being confidently inferred.
- Explicit uncertainty and failure states.
- Product artifacts alongside the code: epics, user stories, and acceptance criteria.
- Clio REST/OAuth2 synchronization plan.
- End-to-end runs on synthetic document packets.
Current boundary
Live Clio OAuth is pending. This is not a production Clio integration. The attorney did not purchase the product, and this should not be represented as a paid engagement or customer conversion.
Why it belongs on a product site
The strongest evidence is the sequence: discovery falsified the premise → the workflow changed → the product was rebuilt → consequential output remained evidence-gated. It is the same product instinct I used in enterprise SaaS, but carried all the way into working software.
Good discovery is allowed to destroy the thing you hoped to build.