Inspiration

Medical prescriptions, diagnostic reports, and discharge summaries are often dense, handwritten, or filled with intricate clinical jargon. Patients frequently struggle to comprehend their medication schedules, critical findings, and physician instructions. DocuSense bridges this gap by providing an accessible, multimodal document explainer that deciphers uploaded medical records into plain, structured insights in seconds.

What it does

• Multimodal Visual Extraction: Ingests images of prescriptions, lab panels, and clinical documents via a responsive Streamlit interface.
• Key Findings Summarization: Distills complex diagnostic statements into easily understandable summaries.
• Medication & Dosage Schedules: Extracts listed medications, intended frequencies, dosages, and usage directions into structured outlines.
• Clinical Action Points & Safety Disclaimers: Highlights essential follow-up reminders while enforcing medical disclaimers to ensure safe patient literacy.
• Dynamic Model Fallback: Built with adaptive model resolution across Google GenAI endpoints to ensure continuous uptime.

How we built it

• Frontend Framework: Streamlit for reactive document uploads and real-time inference streaming.
• Vision & Intelligence Engine: Google GenAI SDK powered by Gemini multimodal vision endpoints for optical document understanding and contextual extraction.
• Image Processing: Python Imaging Library (Pillow/PIL) for pre-processing uploaded document media.
• Deployment & Version Control: Structured, open-source Git and GitHub repository.

Challenges we ran into

• Handling varying resolutions, formats, and artifacts across uploaded medical images.
• Resolving API model identifier deprecations dynamically without breaking the live application runtime.

Accomplishments that we’re proud of

• Delivering a functional, zero-latency clinical document translator that runs locally with clean multimodal reasoning.
• Achieving structured parsing of pharmaceutical schedules directly from raw document snapshots.

What we learned

• Navigating multimodal prompt architectures for specialized domain literacy.
• Handling resilient API failovers and client configuration patterns in Streamlit.

What’s next for DocuSense

• Support for multi-page PDF diagnostics and historical trend comparisons.
• Integration of regional multilingual translation to support vernacular healthcare access across diverse patient populations.