AI Agent & RAG Developer — Production-Ready Automation System (Claude API, LangChain, n8n, FastAPI)

Overview We're a growing B2B services company looking for an experienced AI Automation Engineer to design and build a production-grade AI assistant and workflow automation system. This is a real production build for a live business — not a prototype, not a demo. If your experience is limited to tutorials or chatbot demos, this isn't the right fit. What You'll Build 1. RAG-Powered Knowledge Assistant Build a Retrieval-Augmented Generation (RAG) pipeline over our internal knowledge base (PDFs, Word docs, Google Drive files, Notion pages) Document ingestion, chunking strategy, embeddings, and semantic search using a vector database (Pinecone or pgvector — recommend one and justify it) Grounded, hallucination-free answers with source citations — the bot must say "I don't know" when the answer isn't in the documents Multi-turn conversation memory and clean escalation to a human agent with full context Accessible via a web chat UI (React/Next.js) and WhatsApp Business API 2. AI Agent Layer with Tool Use LLM orchestration using LangChain or LangGraph (multi-step tool calling) Function calling / structured JSON outputs (Pydantic schema enforcement) for CRM updates, appointment booking (Calendly/Google Calendar), and lead qualification (Hot/Warm/Cold scoring) Human-in-the-loop approval workflow — no AI output touches a customer without review 3. Workflow Automation n8n (or Make.com) workflows connecting Gmail/Outlook, HubSpot CRM, Slack, Airtable, and Google Sheets Email triage: classify incoming emails, draft AI-generated replies for team review (no auto-send), and log everything Webhooks, REST API integrations, OAuth 2.0, error handling, retries, and rate-limit management Tech Stack AI: Anthropic Claude API (preferred) and/or OpenAI API — model-agnostic layer so we're not locked in Backend: Python, FastAPI, async, Celery, PostgreSQL + pgvector, Redis Frontend: React / Next.js, TypeScript, Tailwind Infra: Docker, Supabase, cloud deployment (AWS/GCP/Vercel) Automation: n8n, Make.com, or Zapier Bonus: Claude Code, Cursor, or similar AI-assisted development tools; MCP servers; voice AI experience (Vapi, Retell AI, Twilio, ElevenLabs) Requirements Proven experience shipping production LLM applications (RAG pipelines, AI agents, tool calling) that real businesses use daily Strong prompt engineering skills — consistent, low-variance structured outputs across hundreds of runs Experience building guardrails against hallucination and prompt injection Comfortable working inside an existing codebase, not just greenfield Clean, well-documented, maintainable code — a non-technical owner must be able to maintain this after handoff Clear written English and async communication Nice to Have HIPAA/GDPR-compliant or regulated-industry AI experience Voice agent builds (inbound/outbound calling, STT/TTS pipelines) E-commerce/Shopify or CRM automation background Fine-tuning or self-hosted open-weight model experience (Ollama, vLLM) To Apply, Please Include: 2–3 examples of production RAG or AI agent systems you've built — what they did, the stack, and the measurable result Your approach to chunking/retrieval strategy for a mixed-format knowledge base How you keep LLM output format-consistent and prevent hallucinated answers Your proposed timeline and milestone structure Start your proposal with the word "Grounded" so we know you read this We're not looking for the cheapest option — we're looking for someone who can build something that works and explain what they built. Strong applicants get a fast response, and successful delivery leads to ongoing long-term work.