Projects I've worked on, from AI agents to distributed systems.
AI agent that rewrites resumes to match job descriptions without keyword stuffing. Built multi-step workflow with Crew AI that extracts role requirements, analyzes tone, and rewrites experience while maintaining narrative consistency. Outputs tailored PDF/Word docs that pass ATS filters. Deployed on GCP with MongoDB for resume version management.
MCP server connecting live F1 telemetry to LLMs for natural language race queries. Processes high-frequency data streams with FastMCP and HttpX, filtering noise to extract meaningful signals like lap times, tire degradation, and pit windows. Enables conversational queries about race strategy and pace without switching dashboards.
Reinforcement learning simulation using SARSA to optimize traffic light timing based on real-time queue lengths. Modeled intersection dynamics with NumPy and Pygame, training agent to learn adaptive switching policies instead of fixed schedules. Reduced average wait times compared to static timers, demonstrating RL for real-time decision-making systems.
ReactNext.jsPostgreSQLPineconeCrew AI
Hybrid search RAG system combining vector similarity and keyword filtering for technical documentation retrieval. Built with Next.js and React frontend, PostgreSQL for metadata, and Pinecone for vector search. Crew AI generates answers with citations, balancing semantic understanding with exact term matching to reduce time spent manually searching PDFs.