LexLayer: from a pile of documents to a reviewed, ready matter.
Production, early users
Lawyers lose hours to intake: reading documents, pulling out key facts, filing them to the right matter, and tracking which stage each matter has reached. LexLayer gives every matter a workspace, extracts text from uploaded documents, and moves work through explicit states that are reviewed and recorded.
- Upload
- OCR and extraction
- Human review
- READY
Constraints
- Client confidentiality comes first.
- Scanned, low-quality documents are normal.
- Lawyers won't rely on a result they can't trace back to its source.
- Every change needs a record of who made it and when.
Key engineering decisions
| Decision | Alternative | Why | Trade-off accepted |
|---|---|---|---|
| Each matter moves through an explicit state machine, ending in READY | Free-form status fields | Predictable workflows, an auditable history, no lost matters | New workflows have to be modelled up front |
| OCR and extraction run as an asynchronous pipeline | Process synchronously on upload | Large scans don't block the screen, and failed jobs can retry | Progress and failure states have to be shown to the user |
| A person reviews and approves extracted results, and the approval is stored | Apply AI output automatically | Lawyers check before relying on output, which builds trust | A human step stays in the loop by design |
| Workspaces and access are scoped to each matter | One shared document pool | Confidentiality between matters and clients | A more involved permission model |
| The workflow doesn't depend on one model provider | Tie the workflow to one provider | Hosted, private or local models can be swapped in to suit confidentiality needs | An adapter layer to maintain |
Failures the system handles
- Unreadable scans are flagged for manual handling, never guessed silently.
- Extraction below the confidence threshold goes to review.
- Duplicate uploads.
- Invalid state transitions are blocked.
- OCR workers that fail are retried.
Results
In production with two practising advocates. [Measured figures to add: documents processed, intake time before and after, extraction accuracy on a sampled set]
How this applies to you
If your team re-keys documents into a case, claim or shipment system, LexLayer's pattern (intake, extraction, review, then a state change) is one we can adapt to your workflow.
All code is private. We offer code walkthroughs under mutual NDA.