Nuvro Care Intelligence

A family should maintain one evolving care story and generate the right view for each professional—not rebuild the story at every appointment.

A multilingual care-intelligence platform that turns a fragmented autism journey into a coherent, role-aware case narrative.

At a glance

• Status: MVP · Active development
• Architecture focus: Healthcare
• Primary outcome: One story, role-aware views
• Stack: Next.js, RAG, Prisma, Postgres, pgvector, Hinglish UX

The challenge

Parents repeatedly retell the same complex history to doctors, therapists, schools, and other care professionals.

The architecture decision

Maintain one longitudinal master case brief and generate grounded, audience-specific views instead of duplicating the source of truth.

System architecture

Responsibility moves through explicit boundaries. Each layer has one primary job; policy, health, and trace signals span the system.

Open the standalone architecture and workflow diagrams.

Architecture layers

1. Guided intake — Capture the child’s story progressively in English, Hindi, or Hinglish.
2. Longitudinal profile — Keep history, strengths, concerns, interventions, responses, and goals together.
3. Evidence layer — Retrieve approved care knowledge with source, topic, language, and audience metadata.
4. Master case brief — Synthesize one structured, traceable narrative without replacing clinical records.
5. Role-aware views — Generate focused briefs for doctors, therapists, schools, and family discussions.
6. Progress loop — Track meaningful changes, unresolved questions, next reviews, and relapse signals.

How the system works

1. Understand the child — Dynamic questions build a structured picture without forcing a family through one giant form.
2. Build one narrative — The platform joins history, observations, interventions, and goals into a master case brief.
3. Ground guidance — Retrieval supplies relevant approved context; generated text remains linked to source material and profile facts.
4. Share the right view — Each professional receives the relevant subset while the family preserves one source of truth.

Architecture decisions and trade-offs

ADR-01: Master brief, not many documents

Audience briefs are projections of one longitudinal record.

Trade-off: The core schema requires more care, but contradictions and repeated storytelling are reduced.

ADR-02: Guidance is not diagnosis

The system organizes history, education, questions, and follow-up; clinical decisions remain with qualified professionals.

Trade-off: The product intentionally avoids some high-risk automation in exchange for a safer and clearer role.

ADR-03: Language is part of architecture

Meaning, source grounding, and structured fields are preserved across English, Hindi, and Hinglish experiences.

Trade-off: Multilingual evaluation adds work, but accessibility is not postponed to a translation layer.

Production qualities

• Safety: Clear scope, uncertainty language, professional escalation, and no unsupported diagnostic claims.
• Privacy: Least-privilege access, encrypted storage, export controls, consent, and sensitive-field isolation.
• Grounding: Retrieval filters by audience and topic; outputs preserve evidence links and profile provenance.
• Continuity: Every brief derives from the same timeline, goals, open questions, and intervention history.

Outcomes

• A single longitudinal care narrative
• Role-specific briefs without duplicated records
• More prepared professional conversations
• A foundation for progress and relapse tracking

Technology stack

• Next.js
• RAG
• Prisma
• Postgres
• pgvector
• Hinglish UX

Related writing

• One master case brief is better than five disconnected summaries
• Evaluating an enterprise AI assistant beyond answer accuracy