APOLLO AI

Landing Page
Patient Profile
Next Steps

Inspiration

Our project was inspired by the challenges healthcare providers face in quickly accessing reliable, peer-reviewed medical knowledge at the point of care. Doctors often need to cross-reference clinical notes, symptoms, and diagnostic findings with the latest research, but this process is slow and fragmented. We wanted to build a tool that acts as a bridge between raw patient data and actionable insights — helping clinicians move from diagnosis to next steps seamlessly.

What We Learned

Throughout the development process, we learned:

How to structure unstructured medical notes into machine-readable JSON.

How to integrate with external biomedical APIs like PubMed (Entrez API) and ClinicalTrials.gov's Data API to fetch relevant studies and ongoing trials.

The importance of user-centric design: clinicians prefer clarity and brevity, not long technical reports.

How MCP servers can enhance medical workflows - implementing Redis-based session management for secure, multi-device patient data persistence.

How We Built It

Data Modeling – We created a schema for medical case notes in JSON, capturing patient demographics, symptoms, medical history, imaging results, treatments, and discharge plans.

NLP & Retrieval Layer – Using a retrieval-augmented generation (RAG) approach, we mapped key fields (e.g., "diagnosis tests", "symptoms") into queries for PubMed and ClinicalTrials.gov.

APIs & Integration –

PubMed via Entrez E-utilities for peer-reviewed articles
ClinicalTrials.gov Data API for ongoing or completed trials
Redis MCP Server for enterprise-grade session management and patient data caching

Frontend – A React-based app with a clean medical-inspired UI

Backend – Node.js/Express to handle ingestion, query generation, and API aggregation.

AI Layer – A SLM fine-tuned on augmented clinical notes to summarize findings and propose evidence-based next steps.

Challenges We Faced

Data sensitivity: Working with clinical data meant ensuring all examples were synthetic or anonymized.

Ambiguity in medical language: The same condition might be described differently across notes (e.g., "brain tumor" vs. "glioma"), requiring normalization.

Session management complexity: Balancing security requirements with user experience across multiple devices and hospital systems.

Outcome

We built a prototype that converts a structured or semi-structured doctor's note into:

A concise patient summary with persistent cross-device sessions.

A list of relevant peer-reviewed articles.

A feed of clinical trials matching the condition.

Suggested next steps (diagnostics, treatment considerations, follow-up).

Enterprise-ready architecture with Redis MCP server enabling secure, scalable deployment across hospital networks.

This project gave us a better appreciation of the complexity of medical decision-making and showed how AI tools can augment — but not replace — human expertise.