AI Chatbot & Customer Support Automation — Mortgage Website

# Overview We are building an AI-powered chatbot platform for a mortgage lending business. The chatbot will serve as an intelligent customer support and lead engagement tool — embedded on the website, connected to the business's CRM and loan systems, and accessible across SMS and email channels. We need an experienced developer to build and deploy this end-to-end. This is a project-based engagement with a project manager overseeing requirements and coordination. Clear communication and the ability to work independently between check-ins is important. --- ## Existing Stack (must integrate with) - Brevo — CRM. Borrower contact data, pipeline stages, and loan status all live here. The chatbot must read from and write to Brevo via its REST API. - POS (Point of Sale) system — where borrowers fill out mortgage applications. Data flows from POS → Brevo. - LOS (Loan Origination System) — tracks active loans and processing status. Data flows from LOS → Brevo. The POS/LOS → Brevo data pipeline is already set up by another contractor. Your job is to connect the AI layer to Brevo — you will not need to touch the POS or LOS directly. --- ## What Needs to Be Built ### 1. AI Chatbot Engine The core brain behind everything. Must: - Integrate with at least one major LLM provider (OpenAI, Anthropic, or similar) - Be designed so the underlying model can be swapped without rebuilding everything (e.g. switching from GPT to Claude should not require a rewrite) - Maintain conversation context across messages - Be grounded in a company knowledge base (RAG system using existing knowledge base — details provided at kickoff) - Use borrower data pulled from Brevo as real-time context so it can answer questions about a specific user's loan or application status ### 2. Website Chat Widget - Embeddable widget that can be dropped into any website via a script tag - Persistent session handling — conversations must not reset on page refresh or tab switch (a previous implementation had this bug) - Connects to the chatbot backend in real time ### 3. Omnichannel Support (SMS & Email) - When a borrower texts or emails the business, the same AI engine continues the conversation - The bot must have access to the same borrower context (from Brevo) and the same conversation history regardless of channel - The system must correctly link an SMS or email message to the right borrower contact in Brevo ### 4. Live Form-Fill Assist - While a borrower is actively filling out the mortgage application form on the website, the chat widget can access the partial form state - The bot uses this live context to answer mid-form questions (e.g. "what does this field mean", "do I need this document") - Requires coordination with the website/form implementation to ensure field-level data is accessible to the widget ### 5. Admin Dashboard & Human-in-the-Loop - An internal dashboard for the business owner/admin to view active and recent conversations in near real-time - Ability to toggle any conversation from AI to human — the admin can take over and type directly to the borrower through whichever channel they're on (web, SMS, or email) - Ability to hand the conversation back to the AI after the human is done - Does not need to be real-time WebSocket level; near-real-time (polling-based) is acceptable --- ## What We Are NOT Looking For - Anyone to build or redesign the marketing website — that is handled separately - Backend infrastructure setup or DevOps — that is managed by the project manager --- ## Ideal Candidate - Has built AI chatbot or LLM-powered applications before, ideally in a customer support or lead qualification context - Comfortable working with third-party REST APIs (Brevo's API is well-documented) - Experience with multi-channel messaging (SMS via Twilio or similar, email inbound/outbound handling) - Can build embeddable JS widgets that work across different website stacks - Able to write clean, maintainable code — this system will need to be handed over and supported after delivery Please include in your application: examples of chatbot or LLM-integrated projects you've shipped, your approach to the multi-channel identity problem (linking a web session, an SMS number, and an email address to one borrower), and your estimated timeline for this scope.