Ankur M.

AI and full-stack engineer. 376 contracts delivered, $1M+ earned, Expert-Vetted (top 1% on Upwork). I build production RAG pipelines, AI agents and LLM applications on Python, Node.js and React, with US and UK overlap hours. 15+ years building enterprise web applications and production AI systems across fintech, healthcare, SaaS, logistics, real estate and edtech. 272 client reviews. Top Rated Plus. WHAT I BUILD Generative AI and retrieval-augmented generation. RAG pipelines grounded in your own documents, policies and product data, using Pinecone, Weaviate, Qdrant and FAISS, with hybrid search and OCR document extraction. Where a smaller machine learning model is cheaper and more predictable than an LLM call, I train one on your data for classification, ranking and recommendation. AI agents and agentic workflows. Multi-agent orchestration with LangChain, CrewAI, AutoGen and LlamaIndex, function calling, tool use and MCP servers, built on ChatGPT and OpenAI models, Claude and Gemini. Prompt engineering and evaluation harnesses so behaviour stays stable as models change. AI chatbots and conversational systems. Customer-facing assistants with guardrails, confidence checks and human escalation paths, wired into CRM and support workflows. MERN stack applications. MongoDB, Express, React and Node.js end to end, from schema design and API layer through to the interface. Multi-tenant SaaS, marketplaces and dashboards, with TypeScript throughout, WebSockets for real-time features, and REST and GraphQL APIs built to scale. Full-stack and SaaS platforms. React, Next.js, TypeScript, Node.js, NestJS, Python, FastAPI and Django. Multi-tenant architecture, subscription billing, role-based access and analytics dashboards. Cloud and data infrastructure. PostgreSQL, MongoDB, Redis, Docker, Kubernetes, AWS Lambda, GCP, CI/CD and event-driven queues. RECENT WORK Sales engagement platform. AI workflows with Twilio VOIP, Mailchimp, Outlook and GSuite integrations driving automated customer engagement at scale. AI SaaS and voice platform. Conversational AI, intelligent workflows and API-driven customer experiences. Healthcare AI diagnostics and patient portal. HIPAA-compliant voice, chat and OCR triage handling intake, symptom analysis, scheduling and automated EMR updates. Intake processing time down 80%. Python, FastAPI, Gemini, Pinecone, React, AWS. Fintech document intelligence and KYC engine. Automated pipeline parsing 100+ page financial statements, tax forms and loan applications into structured PostgreSQL records with full audit-trail logging. Human audit overhead down 65%. Claude, Document AI, Node.js, PostgreSQL, Docker. Real estate AI search. Hybrid RAG and vector search letting buyers query properties in natural language and by image, with automated lead routing into HubSpot. React Native, LangChain, Weaviate, Node.js. Logistics routing and order assistant. Real-time driver tracking, automated route optimisation and an MCP-powered assistant letting warehouse teams query live inventory and orders in natural language. Next.js, Python, Redis, Google Maps API. Full project details and links are in my portfolio below. HOW I APPROACH AI SYSTEMS The LLM is the reasoning and drafting layer. Business rules, permissions, pricing and anything that writes to a system of record stay in deterministic application logic, behind explicit approval where the stakes justify it. That separation is what makes an AI feature safe to put in front of customers, and it is the difference between a demo and something that runs unattended for a year. Every build ships with logging, monitoring, error handling, tests and an auditable trail. WORKING WITH ME I work US Eastern and UK overlap hours and reply within the business day. One point of contact from architecture through to deployment: schema design, backend, LLM pipeline, frontend and cloud. Clear milestones and weekly progress. Best fit for founders, CTOs and product leads who need a system that holds up in production rather than a prototype, and who want the architecture reasoning explained rather than hidden. Send me your project goals, current stack and target timeline and I will come back with the architecture I would recommend and where the risks sit.