Prabhakar Punj

Focused on advancing my software engineering expertise into agentic AI, building and deploying production-ready applications with LangGraph, LangChain, and modern cloud infrastructure. Applied my prior backend and distributed-systems experience to multi-agent architectures, RAG, and LLM-powered workflows to create the following SaaS products.

LoopHolio is an AI-powered Terms of Service analyzer that turns lengthy legal documents into structured, clause-level risk reports. It uses a multi-agent LangGraph pipeline for clause extraction, risk detection, legal reasoning, and plain-English explanations, while preserving traceability back to the original source text. Built with Next.js, FastAPI, LangChain, Pydantic, and IBM Docling for document processing and reliable structured outputs.

Built an 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.