Adam Kamel

Built the document intelligence and financial RAG platform behind a restructuring analytics product

• Replaced a paid parsing service with an open-source pipeline, saving $25K with $350K projected at scale
• Built a financial RAG system on a PostgreSQL vector store, retrieving evidence across 250K+ documents
• Shipped real-time distress alerts from filings and 20 insolvency trustees, saving analysts 5-10 hours weekly
• Built a survival model predicting one-year insolvency risk for public companies with 80% accuracy

Leading WorldFold, a project exploring world models and reinforcement learning for robotic cloth manipulation

• Co-leading 8 ML researchers and engineers building world-model RL control for a dual-arm cloth-folding robot
• Overseeing a custom MuJoCo cloth simulator integrating SO-101 robot arms with deformable cloth physics
• Guiding development of a Dreamer-style world model that predicts how cloth responds to each robot action

Trained statistical and deep learning classifiers to identify market regimes in noisy financial time series

• Applied statistical and deep-learning models to identify market conditions across 16 years of financial data
• Expanded the training dataset with a rolling evaluation, improving accuracy from 42% to 86%
• Designed HMM state and transition modeling to capture regime persistence and shifts without labeled data

Developed AI agent integrations and improved natural language processing systems

• Improved natural-language endpoint routing accuracy by 28% with multi-step vector search over text embeddings
• Developed FastAPI integrations for 30+ external services
• Wrote Pytest cases covering 200+ integrated service endpoints