Anurag S.

I’m an AI developer with $700K+ earned, 9,200+ hours, 130 completed jobs, and Top Rated Plus status on Upwork. I build practical AI products that solve real business problems, not just demos that look good for a few minutes. My work covers AI agents, LLM chatbots, RAG systems, voice AI, OCR, document intelligence, AI search, workflow automation, and AI features inside web and mobile products. I work with OpenAI, Claude, Gemini, LangGraph, LangChain, LlamaIndex, MCP, vector databases, and modern cloud infrastructure. But I don’t start with a model or framework. I start with your workflow, data, users, and the outcome you actually need. What I can help you build • AI agents that connect with your APIs, CRM, databases, internal tools, and third-party platforms using LangGraph, OpenAI Agents SDK, Claude tool use, and MCP • RAG chatbots and knowledge assistants that answer questions from your documents, website, knowledge base, internal policies, product data, or private company information • Voice AI agents for inbound and outbound calls using Vapi, Retell, Twilio, ElevenLabs, and custom backend workflows • OCR and document AI for invoices, PDFs, forms, handwriting, document classification, extraction, tagging, and structured data processing • AI features inside existing SaaS, web, or mobile products, including summarisation, classification, AI search, recommendations, content generation, image generation, and workflow automation • Internal AI tools for teams, including custom GPTs, Claude Projects, prompt systems, evaluations, retrieval workflows, and AI-assisted business processes A few examples of recent work One long-term project was an AI appointment chatbot that grew into a 3,200+ hour engagement. The work went far beyond adding a chatbot. It involved building the conversation flow, connecting business logic, handling user intent, integrating backend systems, and improving the product over time based on real usage. I also worked on an AI storybook platform where users can create personalised stories using AI. The product combined GPT-4o, Supabase, Stripe, user accounts, generated content, and a workflow that takes the user from an initial prompt through to a finished book. Another project involved building an AI call-review system using Fathom and GPT. It analyses sales calls, scores conversations against defined criteria, identifies important moments, and helps teams review calls without manually listening to every recording. I’ve also worked on OCR and document-processing systems that automatically tag PDF documents, extract structured information from forms, and convert handwritten content into usable digital data. For enterprise and internal AI projects, I’ve worked on solutions where AI needs to operate within existing workflows rather than as a standalone chatbot. That usually means connecting the model with databases, APIs, reporting systems, permissions, business rules, and human review steps. My typical stack Python, FastAPI, Node.js, LangChain, LangGraph, LlamaIndex, OpenAI, Anthropic Claude, Gemini, Pinecone, pgvector, Supabase, PostgreSQL, AWS, Azure, Twilio, Vapi, Retell, and ElevenLabs. How I usually approach a project I normally start by understanding the workflow before suggesting the architecture. What data does the AI need? Where does that data live? What should the AI be allowed to do? What happens when it is uncertain? Which actions should require human approval? These questions matter much more than simply choosing the newest model. For larger AI products, I prefer proving the core workflow on real data first. Once we know the approach works, I move into the production build, including API integrations, authentication, logging, monitoring, evaluations, error handling, cost controls, and deployment. I also pay attention to areas that are easy to miss in AI projects, such as hallucination control, prompt versioning, fallback behaviour, permissions, retrieval quality, token usage, latency, and model cost. A good AI product needs to be useful and reliable after launch, not only during a demo. If you already know exactly what you want to build, I can help with architecture and development. If you only have the business problem and are still figuring out whether you need an agent, RAG system, voice AI, OCR workflow, or a simpler automation, I can help with that too. Share your use case, current workflow, and the tools or data involved. I’ll help you work out a practical architecture, the right model choices, and what it will realistically take to build and run.