AI Engineer / Full Stack — MCP Server, RAG, pgvector, Multi-Agent LLM Gateway for Deal Finder SaaS

We are hiring selected developers to build a production AI layer for CardPerks, a live card-deals marketplace (web + React Native). This is not a toy chatbot. We need grounded AI architecture: MCP tools, RAG over real deal data, multi-agent workflows, and a cost-aware LLM gateway so customers can find deals for free. What we are building An AI-powered deal discovery and content automation system on top of our existing SaaS: MCP server (Model Context Protocol) with structured tools for deal search, deal detail, draft generation, duplicate checks, media prep, and social publish actions RAG (Retrieval-Augmented Generation) grounded in our deal catalog, merchant/city/bank context, brand voice rules, and prior social posts PostgreSQL + pgvector as the relational database and vector database for embeddings, semantic search, and retrieval Multi-agent system with tool calling / function calling, permission gates, and audit-friendly tool results LLM gateway / model router that can switch between free-tier LLM APIs, local Ollama, and paid providers (OpenAI, Anthropic Claude, Gemini) without rewriting the product Graceful degradation so if AI quota is exhausted, classic PostgreSQL filters and deal browsing still work AI agents in scope Deal Finder agent (customer-facing) — natural-language deal Q&A using RAG + pgvector + MCP search tools. Default browse/filter stays PostgreSQL with no LLM cost. Social Publisher agent — grounded Facebook/Instagram draft and publish flow via MCP tools on our Graph API automation. Deal Ops agent — internal catalog hygiene and publish-queue assistance with bounded MCP actions. Cost and free UX requirements Customer deal discovery must stay free by design: City / bank / merchant / card filters with no LLM cost AI chat uses free-tier APIs first, then Ollama, with optional OpenAI/Claude fallback Answer caching, small RAG contexts, and daily AI caps Paid models are backup only, not a hard dependency Existing product stack you will extend Laravel REST APIs (deterministic deal and publish backbone) Angular web + React Native mobile on shared APIs Facebook Graph API / Instagram Graph API publishing with idempotent jobs Migration path to PostgreSQL + pgvector for AI retrieval (do not invent deal prices or eligibility in the LLM) Core business rules stay in the API/database. The AI retrieves and acts through tools. It does not replace the source of truth. Who we want We are inviting a small number of strong freelancers. Ideal fit: Proven work with MCP servers, RAG pipelines, vector databases (pgvector), and tool-calling agents Strong PostgreSQL, API design, and production backend skills (Laravel/PHP and/or Node/TypeScript) Experience with OpenAI / Claude / Gemini / Ollama integration and prompt engineering Comfortable with cost-aware AI gateway design, caching, and rate limits Clear async communication and clean handoff documentation Bonus: React/Angular chat UI, React Native, social Graph API automation, SaaS multi-tenant patterns. Deliverables Architecture short note (MCP tools, RAG index, agent boundaries, model router) Working MCP server with agreed tools PostgreSQL + pgvector RAG retrieval for deals and knowledge docs Deal Finder agent wired to web (and optionally mobile) LLM gateway with free-tier + Ollama + paid fallback switching Basic evals, logging/audit of tool calls, and a short handover guide Optional stretch: Social Publisher agent and Deal Ops agent To apply Please include: Links to 1–2 similar projects (MCP, RAG, agents, or pgvector) Your proposed approach for free-tier + Ollama + OpenAI/Claude routing Timeline and fixed-price or milestone quote for MVP (Deal Finder + MCP + RAG + gateway) Confirm you can work with our existing Laravel API and keep deal data deterministic We will only interview selected applicants.