Musharib Nadeem

Summary

What I did

• Built an OCR pipeline (FastAPI, Docker, Redis Streams) auto-populating billing data at 90%+ accuracy.
• Designed a multi-agent coding platform with hierarchical RAG and a rules engine at ∼80% accuracy.
• Shipped LiveKit voice agents for booking and follow-ups using LangGraph to manage complex states, handling hundreds of calls/day.
• Trained XGBoost/PyTorch models on parsed EDI 837 claims: 80%+ accuracy, catching 65% of all denials.
• Trained in-house ML models to predict whether a denial would occur on a particular medical claim.
• Fine-tuned a model for correct diagnosis detection and medical language understanding.

Results

• Achieved 90%+ accuracy in auto-populating billing data via OCR pipeline.
• Reached ~80% accuracy on a multi-agent medical coding platform.
• Successfully handled hundreds of calls per day with LiveKit voice agents.
• Caught 65% of all denials using a multi-model ML pipeline with 80%+ accuracy.