Anurag Jain — Building ML Systems That Ship

Research in AI/ML with focus on deep learning, reinforcement learning, and generative models.

• Implemented transformer-based RL agents for Skat (imperfect-information card game), achieving 26x improvement in move prediction accuracy (3% → 80%) through behavioral cloning
• Engineered custom MLP policy networks with logit-level action masking, reducing training loss by 58%
• Built hybrid Java-Python architecture integrating game engine with ML models for real-time decision-making

Built full-stack ML learning platform serving 40+ active users with automated code evaluation and real-time feedback.

• Developed browser-based IDE with secure Dockerized sandbox for Python execution, implementing isolated environments for safe user-submitted code
• Deployed on GCP with CI/CD pipelines (GitHub Actions), automating container orchestration for 30+ ML problems across 9 topics
• Built automated grading system evaluating ML implementations with instant visual feedback (metrics, plots)

Built ML models for analyzing complex BIM data to automate parts of construction planning.

• Structured high‑dimensional building information (BIM) for downstream ML tasks.
• Achieved 98.21% training accuracy and 95.17% testing accuracy, with robust F1-scores (Train F1: 0.9239, Test F1: 0.8113).

Built a cross‑platform app in Flutter/Dart and implemented digital authentication flows.

• Shipped UI features and improved reliability of auth and session handling.
• Designed user-centric interfaces in Figma, improving UX for cross-platform mobile applications.