🩺 MediPrep — Smarter, Stress-Free Doctor Visits

Track - Healthcare and Wellness

Team

Trusha Neogi Indrayudd Roy Chowdhury Nithishkumar Sivakumaran

💡 Inspiration

Doctor visits are often overwhelming — patients forget to mention symptoms, bring up concerns, or recall what was said.
MediPrep was born from this everyday frustration: giving patients a way to stay organized, confident, and informed — without relying on the internet or compromising privacy.

🧠 What We Learned

We learned how to combine AI-driven intelligence with full offline privacy, exploring how to keep everything on-device while still being useful.
We also deepened our understanding of lightweight LLM deployment, structured healthcare data capture, and cross-platform CLI automation for seamless medical documentation workflows.

🛠️ How We Built It

We developed MediPrep as a CLI-first mobile prototype in under 5 hours, built entirely using:

• 🧩 Codex + Gemini CLI for fast iteration and modular, text-based UI development
  
• 💬 Gemma-3-1B-IT — an open, compact large language model that runs locally to power offline question generation, summarization, and document processing
  
• 📂 JSON to store visits, folders, transcripts, and metadata
  
• 🎙️ Local audio capture and transcription for secure, offline medical note-taking
  
• 🧱 Text-based navigation and command handling to simulate mobile UI flow within the CLI
  

The app experience:

1. Start with a main folder view of illnesses and past visits
   
2. Add a new or follow-up appointment using a simple form
   
3. Upload or attach prescriptions and supporting files
   
4. Run on-device AI (Gemma-3-1B-IT) to generate relevant, illness-specific questions
   
5. Record answers and generate a visit summary locally — no internet, no data sharing

⚙️ Why Gemma-3-1B-IT

We used Gemma-3-1B-IT, a lightweight instruction-tuned model optimized for efficiency, to:

• Process medical notes and files entirely offline
• Generate context-aware doctor questions from structured patient input
  
• Summarize recordings into concise, human-readable summaries
  This ensures data never leaves the device, aligning with the privacy-first nature of MediPrep.

⚙️ Challenges

• Running Gemma-3-1B-IT locally while maintaining real-time responsiveness
  
• Designing a multi-screen workflow entirely in the CLI
  
• Managing complex folder and file hierarchies with JSON persistence
  
• Ensuring stable audio recording, saving, and playback simulation without GUI dependencies
  

🚀 Outcome

MediPrep empowers patients to approach doctor visits like pros — informed, organized, and confident — all while keeping their health data private and local.
By integrating Gemma-3-1B-IT, we demonstrated that powerful AI doesn’t need to compromise privacy — it can live securely on your device, working for you, not against you.

Built With

• codex
• figma
• flutter
• gemini
• gemma
• vscode
• xcode