Sayan De

AI-Powered Tool Orchestration via Autonomous Agents

• Next.js, TailwindCSS, Express, tRPC, FastAPI, LangGraph, LangChain, LangSmith, MCP, PostgreSQL, Prisma, Docker, AWS
• Architected a full-stack monorepo with a real-time chat interface using Next.js 16 / React 19 frontend and Express/tRPC backend, enabling seamless bidirectional communication.
• Engineered an autonomous AI agent system using LangGraph with Plan → Route → Execute pattern, dynamically orchestrating 20+ MCP tools to execute complex multi-step workflows.
• Reduced LLM token consumption by 90% through intelligent context management, implementing selective tool binding and conversation state pruning.
• Achieved 80% faster response times by implementing smart model routing and prompt compression, dynamically selecting LLMs based on task complexity.
• Decreased hallucinations by implementing a Human-in-the-Loop (HITL) system with state-based approval workflows for safe execution of critical actions.
• Implemented async memory persistence after workflow completion and agent context injection, allowing personalized multi-session interactions with strict user data isolation.
• Ensured 99.9% system uptime by deploying production infrastructure on AWS EC2 with automated CI/CD pipelines and containerized microservices.

Tune LaTeX Resumes & Cover Letter with tuned AI agents

• LangChain, LangGraph, FastAPI, Next.js, PostgreSQL, Docker, AWS Lambda, EC2, ALB, Route53
• Built a self-learning AI agent with long-term memory that adapts to user preferences and preserves context.
• Reduced LLM latency by 80% and costs by 90% through intelligent routing between weak and strong models.
• Optimized LLM for LaTeX code generation using Few-Shot Prompting to ensure syntactic accuracy.
• Decreased token usage by 70% and improved response times via Prompt Compression with LLMLingua.
• Minimized hallucinations in job-application conversations by implementing isolated chat session threads.
• Integrated a LaTeX compiler with live code editor for dynamic document generation.
• Maintained website Core Web Vitals (FCP, LCP, FID, CLS, SEO) with an average score above 95%.
• Deployed backend APIs using a hybrid architecture: LangChain & LangGraph API on AWS Lambda (serverless) for scalability and a LaTeX compiler API on AWS EC2 behind a load balancer for reliability and performance.

Uber's internal on-call copilot (Uber's Engineering Blog)

• LangGraph, LangChain, SambaNova (Llama 3.1), Google Gemini 2.5, FastEmbed, Qdrant, PyPDFium
• Replicated how Uber built their internal Agentic AI chatbot with LangGraph that saved 21,000 developer hours.
• Built an Agentic RAG pipeline using LangGraph, integrating small and large LLMs for query rewriting, hybrid (dense + sparse) retrieval, and context-aware answer generation.
• Optimized document retrieval by storing vectorized data with custom metadata in Qdrant, enabling hybrid search (semantic + BM25) and precise filtering based on metadata conditions.