How ajay-1011-git/cureva works
Cureva is a clinical-study review system: Atlas answers questions over study records and produces evidence-backed findings; later stages review and monitor those findings, with forecasts and drafted paperwork. The web app exposes Atlas, multilingual avatar intake, review, and surveillance. The README documents the major capabilities, while sampled source confirms selected ownership and isolation boundaries; some subsystem wiring is therefore shown only where documented or directly supported.
Main entry points: server.py API server · crew.py Review cycle · atlas.py Atlas · watch.py Period surveillance
Components 24
Study analysis
- Atlas
stage1/atlas.py - Study loader
study.py - Answer schemas
schemas.py - Finding graph
graph/finding_graph.py
Patient intake
- Speech services
intake/sarvam_client.py - Patient context
intake/patient_context.py - Avatar response
intake/groq_client.py - PRO records
stage1/atlas.py
Finding review
- Review cycle
stage2/crew.py - Review memory
stage2/memory.py - Tribunal rounds
tribunal/round1.py - Tribunal model client
tribunal/client.py
Study surveillance
- Period surveillance
stage3/watch.py - Surveillance memory
stage3/memory.py - Deviation forecast
forecast/montecarlo.py - Drafted artifacts
execute/templates.py - Artifact polishing
execute/polish.py
Web interface
- Avatar intake UI
webapp/frontend/src/pages/AvatarTest.jsx - API server
webapp/server.py - Study dashboard
webapp/frontend/src/App.jsx
Other components
- Reviewer
- Study CSVs
- Sarvam speech API
- Groq model API
Connections 27
- Reviewer → Study dashboard: uses
- Reviewer → API server: asks
- Study CSVs → Study loader: loads
- Study loader → Atlas: provides records
- Atlas → Answer schemas: validates answers
- API server → Atlas: dispatches questions
- Atlas → API server: returns answers
- API server → Finding graph: observes findings
- API server → Review cycle: runs cycles
- Review cycle → Atlas: requests detections
- Review cycle → Review memory: records review state
- Review cycle → Tribunal rounds: requests deliberation
- Tribunal rounds → Tribunal model client: requests model calls
- Tribunal model client → Groq model API: calls
- API server → Period surveillance: runs period walk
- Period surveillance → Review cycle: runs review cycles
- Period surveillance → Surveillance memory: records period state
- Period surveillance → Deviation forecast: requests forecasts
- API server → Drafted artifacts: serves artifact drafts
- Drafted artifacts → Artifact polishing: drafts for polishing
- Avatar intake UI → API server: sends turns
- API server → Speech services: transcribes and speaks
- Speech services → Sarvam speech API: calls
- API server → Patient context: loads context
- API server → Avatar response: requests response
- Avatar response → Groq model API: calls
- API server → PRO records: adds intake records