A lightweight AI-powered prototype that analyses client–coach conversations and generates structured client intelligence for health coaches.
This project was developed as part of the GenAI Product Intern – Client Intelligence Platform assignment for FUME.
https://shreeyabbb-cloud.github.io/fume-client-intelligence/
- Weekly client summary
- Nutrition adherence analysis
- Exercise and step tracking
- Sleep summary
- Water intake summary
- Symptoms and stress detection
- Engagement level assessment
- Key barriers identification
- Pending action items
- Risk and attention flags
- Evidence-grounded insights
- Human review workflow (Approve / Edit / Reject)
Each finding is categorised as:
- ✅ Confirmed Fact
- 👤 Client Reported
- 🤖 AI Inference
- ❔ Missing / Unavailable Information
The prototype is designed to minimise hallucinations by:
- Returning structured JSON output
- Providing supporting evidence for each major finding
- Separating facts from AI inferences
- Avoiding unsupported numerical values
- Explicitly marking unavailable information instead of guessing
- HTML5
- CSS3
- Vanilla JavaScript
- Anthropic Claude API (optional)
- JSON Schema
Open the application and click Load Demo Output.
No API key is required.
Enter your Anthropic API key and analyse a conversation using Claude.
Note: The direct browser API integration is intended only for demonstration purposes. In a production deployment, requests should be routed through a secure backend to protect API credentials.
- Paste the client–coach conversation.
- Generate structured analysis.
- Review AI-generated insights.
- Verify supporting evidence.
- Approve, edit, or reject findings.
- Compile the reviewed report.
index.html
README.md
- Secure backend API integration
- Persistent report storage
- Multi-week trend analysis
- Coach dashboard with client history
- Wearable device integration
- Confidence visualisation
- Export to PDF/EMR
- Authentication and user management
This prototype is intended for demonstration purposes only.
It is designed to assist coaches by summarising conversations and highlighting relevant observations. It does not provide medical advice or clinical diagnosis. Final decisions should always remain under human supervision.