AI Engineer for Customer Chat Assistant

We run an established two-sided marketplace connecting homeowners with craftsmen. We want to add an AI chat assistant to our site that helps homeowners describe the job they need done, answers their questions about the process, and guides them toward posting a project and receiving offers from craftsmen. This is not a generic chatbot bolted onto a webpage. The assistant must be grounded in our own content and data — our service categories, how our marketplace works, and our FAQs — so it gives accurate answers and never invents things. It should feel like a knowledgeable, on-brand helper that moves users toward taking action, not just a Q&A box. Scope of work: Design the assistant's architecture (LLM with retrieval/RAG over our content, guardrails against hallucination, conversation flow) Build the chat service: handle multi-turn conversations, ground answers in our knowledge base, and recognise when a user is ready to post a job and guide them there Implement safeguards — the assistant must stay on-topic, decline confidently when it doesn't know, and never fabricate prices, names, or guarantees Expose it as a clean API endpoint we can integrate into our existing stack Deliver an embeddable chat widget for our site (or integrate with one we specify) Document the approach so our in-house team can maintain, tune, and extend it Our stack (for context): PHP/Laravel, Vue.js, MySQL, MeiliSearch, Anthropic Claude API. You don't need all of these, but your solution must integrate cleanly with a PHP/MySQL backend, and we have a strong preference for the Claude API. You're a strong fit if you: Have shipped at least one production LLM chat assistant with retrieval/RAG (please link it) Know how to reduce hallucination and keep an assistant grounded and on-topic Can design conversation flows that drive a user toward an action, not just answer questions Communicate clearly in English and can overlap with European/Asian business hours To apply, answer these in your proposal (proposals that skip these are ignored): How do you keep an LLM assistant from inventing answers — fabricating prices, guarantees, or facts not in our content? Be specific. Briefly: how would you architect grounding the assistant in our marketplace content and FAQs? Link one comparable chat assistant you've built in production. Engagement: Fixed-scope milestone project to start (a working proof of concept grounded in a sample of our content), with potential for an extended build if the first milestone holds up. Please include your estimated timeline and budget for the proof-of-concept phase.