Anurag S.

I've completed 126 projects on Upwork, logged 9,000+ hours, and earned over $600K here. Most of my recent work has been with RAG, AI agents, and voice applications that use private data and connect with business systems.

One booking chatbot grew into a 3,200-hour engagement. I worked on appointment flows, changes, follow-ups, and the steady stream of improvements that come after a product is in use. I also spent 1,315 hours on a personalized storybook platform built with GPT-4o, Supabase, and Stripe. Other projects have covered handwriting recognition, health-insurance chatbots, coaching assistants, transcript review, PDF analysis, and voice-agent troubleshooting.

I take on work such as:
• RAG over policies, contracts, case files, and internal knowledge
• LangGraph or LangChain agents with memory, tools, and approval steps
• Voice agents for reception, booking, support, and lead qualification
• Multi-agent workflows for research, review, and structured decisions
• Prompt testing, tracing, evaluation, PII handling, and guardrails
• Debugging live AI products with weak retrieval, slow responses, or unreliable tool calls

On HestaVoice, I worked on configurable voice agents for inbound and outbound calls, campaign scheduling, call history, analytics, calendar booking, and CRM updates. The stack includes Vapi, LiveKit, Deepgram, OpenAI, Claude, FastAPI, React, and Next.js.

INVEXTRA involved six specialist AI advisors, LangGraph and LangChain orchestration, Pinecone retrieval, document uploads, persistent sessions, and automatic PII masking. I worked on the AI layer as part of a larger SaaS product with case workflows, user roles, subscriptions, and administration.

I test with the client's actual documents, questions, and actions as early as possible. When an existing system gives poor answers, I trace the issue through retrieval, chunking, prompts, memory, tool calls, and latency before changing anything.

My main tools are Python, FastAPI, LangGraph, LangChain, OpenAI, Claude, Pinecone, PostgreSQL, Redis, Supabase, Docker, AWS, and Azure. I can join an existing codebase, review a prototype, or help shape a new build. A short note about the data, users, and tools involved is enough for me to suggest a sensible first step.