MedScope

AI-powered literature and evidence intelligence for medical affairs and publication planning.

Built as a proof of concept for ContentEd Med.

What it does

For any of the tracked molecules, MedScope searches PubMed, resolves brand names and synonyms, and classifies papers into four categories requested by medical affairs teams:

1. Clinically Relevant Articles — clinical context, animal/in-vitro data excluded
2. Safety & Efficacy — adverse events, tolerability, efficacy endpoints
3. Trial Results — RCT outcomes with reported results
4. Real-World Evidence — observational, registry, and real-world studies with geographic drill-down

Each category displays a ranked table of papers with a composite score based on journal prestige (SJR) and citation count.

Tech stack

• Streamlit (UI)
• PubMed E-utilities (literature source, no API key required)
• Groq + LLaMA 3.3 (classification and extraction)
• RxNorm (brand/generic/synonym resolution)
• SCImago Journal Rank (journal prestige scoring)

Structure

medscope/
├── app.py              # Streamlit entry point
├── config/             # Editable rules, molecules, prompts, scoring weights
├── core/               # Engine — PubMed, LLM, pipelines (do not modify)
├── ui/                 # Streamlit components (do not modify)
└── data/               # Static lookup files (journal rankings, synonyms)

Running locally

pip install -r requirements.txt
streamlit run app.py

A free Groq API key is required. Get one at https://console.groq.com.

Deployment

Designed to deploy to Streamlit Community Cloud. Connect this repository and set GROQ_API_KEY in the app secrets.

Status

Proof of concept, April 2026. Scope is limited to 10 reference molecules and English-language sources. Cached to a maximum of 50 papers per molecule per category to keep demo latency low.