Messiah Majid

AI-powered insurance prior authorization appeal letter generator that reduces physician appeal writing from 30-60 minutes to under 2 minutes, with RAG over 11 medical guidelines and live PubMed evidence search.

• 5-layer anti-hallucination system ensuring every claim is traceable to clinical notes or guidelines
• RAG pipeline over 11 medical guidelines with live PubMed search via NCBI API
• Multi-model LLM support with automatic fallback (Claude Haiku → Gemini → OpenRouter)

AI-powered drug-target interaction predictor that computationally screens whether a drug molecule will bind to a protein target in under 500ms, trained on 30K+ experimentally validated interactions.

• 0.935 AUROC on Davis Kinase Dataset with XGBoost on 2,483-dimensional feature vectors
• SHAP explainability for every prediction — critical for scientific trust
• Similar drug search using Tanimoto similarity over known compounds

Data-driven analysis tool examining whether prediction markets accurately forecast outcomes, analyzing 600+ resolved markets from Metaculus and Manifold.

• Analyzed 600+ resolved markets with calibration testing
• Built interactive dashboard with FastAPI and Plotly
• Identified favorite-longshot bias: low-probability events overpriced by ~3.5%

Chrome extension that detects social engineering in real-time using fine-tuned LLMs.

• 95% detection accuracy on 10K+ message dataset
• Real-time analysis with sub-second response
• Shipped to Chrome Web Store

Explored computational virology through hands-on work with an open-source capsid modeling package.

• Studied bioinformatics methods and scientific literature
• Gained experience with protein structure analysis pipelines