Prabhakar Punj

Building AI agentic systems with LangGraph and LangChain, including a modular multi-agent research assistant with subgraph architecture, Pydantic-validated state schemas, and checkpointed memory. Applying production engineering practices — clean module boundaries, structured typing, iterative testing — to agentic AI development. Currently deepening expertise in RAG pipelines and evaluation frameworks.

Shipped VoIP, SMS, and voicemail features used by 35,000+ companies, integrating Twilio's API via webhooks and event-driven handlers. Led a monolith-to-microservices migration in Node.js and raised unit test coverage from 60% to 95%. Built Kafka producer/consumer pipelines to offload heavy DB operations and designed a custom SMS rate-limiter on AWS SNS/Redis for reliability at scale.

Cut page load times by 75% by eliminating redundant backend calls and enabling tree-shaking across the component library. Introduced Redux for state management on an inventory system, improving UI responsiveness under heavier data loads. Drove cross-browser compatibility testing and CSS fixes to stabilize the product across environments.

Reduced frontend bundle size by 15% through component refactors and backend API cleanup on a pharmaceutical e-commerce platform. Tuned Node.js/MySQL queries to improve app performance, and built pagination and filtering to handle large product catalogs. Also shipped a C# desktop catalog-sharing tool with a product recommendation API.

Led a research effort to automate detection of software design patterns, cutting manual analysis time by 90%. Built Linux/Java tooling to convert source code into Abstract Syntax Trees and wrote algorithms to extract class relationship data from AST nodes. Prepared labeled datasets and pattern-recognition logic that fed downstream ML models.