Noted, Doctor

Problem: Clinicians spend excessive time on manual documentation during patient consultations—handwriting notes and clicking through healthcare systems—reducing time for actual patient care and diagnosis.

Solution: An intelligent voice assistant that automatically listens to patient-clinician conversations and extracts relevant information according to standard medical documentation schemas, acting as a personal AI scribe for healthcare providers.

Key Impact & Metrics

• Time Savings: Reduces documentation time per appointment by eliminating manual data entry
• Efficiency Gain: Increases patient throughput rate through streamlined workflows
• Focus Shift: Enables clinicians to concentrate on patient care rather than paperwork

User Experience

1. Start: Clinician presses single button to begin recording
2. Converse: Natural patient consultation proceeds uninterrupted
3. Stop: End recording with button press
4. Process: AI agents automatically extract and structure information
5. Review: Formatted notes ready for validation

Technical Implementation

Core Stack:

• DSPy: LLM orchestration with auto-prompt optimization
• FastAPI: Backend API endpoints
• Groq: LLM provider platform
• Mem0ai + Qdrant: Patient session memory management
• React: Frontend interface (built with Lovable)
• Docker: Containerized deployment

AI Models:

• Whisper-V3-turbo (speech recognition)
• Llama-3.3-70b-versatile (decision making)
• GPT-OSS-120b (session memory updates)
• Text-embedding-v3-small (semantic embeddings)

Judging Criteria

Innovation: Replaces manual administrative burden with intelligent AI agents, moving beyond hard-coded solutions to adaptive, learning systems.

Feasibility: Simple two-button interface requires no technical expertise. One-click start/stop functionality makes adoption seamless.

Technical Merit: DSPy framework enables automatic prompt improvement and future model compatibility without code changes. Production-ready architecture.

User Benefit: Eliminates manual data entry, reduces cognitive load, increases patient face-time, and improves documentation accuracy.

Deployability: Rapid deployment via uv + uvicorn + Docker. Scalable architecture ready for hospital system integration, real-time voice APIs, and expansion to other administrative healthcare tasks (lab results, test documentation).

Validation & Next Steps

• Current: Standalone personal assistant for individual clinicians
• Future: Direct integration with hospital EHR systems, real-time processing, multi-department deployment

Solution ideas was brainstormed and decided two days before the event. Wire frames and user flow is roughly drafted before the event. Due to unexpected team up, the original planned svelte front end stack was abandon and wire frames design and user flow was used as a base to build the react frontend. Everything else is created during the hackathon.