Provider flexibility
Server-side adapters support Anthropic Claude and Google Gemini. Deployment configuration selects the provider and model. Browser code never receives model API keys.
TECHNOLOGY
A structured workflow.
A visible evidence trail.
Specialized document extraction, explicit consistency checks and server-side synthesis. Built around the documents in the case.
Controlled PDF ingestion File validation · page limits · session ownership
↓
Policy specialistProvider-backed JSON extraction
Discharge specialistProvider-backed JSON extraction
Bills specialistProvider-backed JSON extraction
Prescription specialistProvider-backed JSON extraction
Test report specialistProvider-backed JSON extraction
↓
Structured case facts + source documentsOriginal PDFs · extracted fields · source-page citations
↓
Deterministic checksMedicine & quantity comparisons · missing reports · dates & line items
Model synthesis & case questionsOriginal PDF context · recent conversation · cited answers
↓
Human review workspacePage viewer · text-location highlights · professional judgment
Document understanding
PDFs are passed to the selected model. Policy tables supplement extraction. Optional Google Vision OCR provides page text for source inspection when configured.
Deliberate scope
This is an orchestrated pipeline of specialized model tasks, not an autonomous adjudicator. The current MVP has no vector search or external legal knowledge base.
Operational boundaries
Case jobs expose high-level stages, not hidden model reasoning. Source access is session scoped. Case storage and deployment retention settings determine data lifetime.
Evidence is not a confidence score.
AI can suggest the wrong page or misinterpret a passage. Exact text matching locates a quote; it does not validate the interpretation. Highlights require searchable PDF text and may be unavailable for scans. The product does not provide calibrated confidence or validated accuracy claims.