AI Developer — Improve AI Assistant in CRM System (Text-to-SQL)

We are looking for an experienced developer to refine and improve the AI assistant in our internal CRM system. This is not development from scratch — the core functionality is already implemented and working. Your task will be to analyze the current implementation, fix existing issues, and propose quality improvements. This is a project with potential for long-term collaboration. We are looking for a developer who is ready to dedicate at least 10 hours per week to this project. About the system: We are building a CRM system based on Supabase (relational database) with a React frontend and a Node.js backend. The system allows users to create their own tables, add fields, and configure relationships using rollups and lookups. Currently, the system contains 10+ objects with real data. Important: the database is shared between two products — the CRM system and a mobile app. The app uses the same database to search for relevant answers to user queries, so the AI assistant must work correctly in both contexts. The AI assistant already understands natural language queries (in Ukrainian) and performs data analytics through SQL generation. Good examples of what we're aiming for: https://crm.lightfield.app/ and https://attio.com/ Technology stack: Frontend: React Backend: Node.js Database: Supabase (PostgreSQL) Hosting: Vercel Architecture highlights: dynamic database structure (users create their own tables and fields), relationships between tables via rollups and lookups, automatic Supabase schema introspection, use of metadata (object and attribute descriptions) as AI context, shared database for both CRM and mobile app. Current issues to solve: High token consumption per query — context delivery needs optimization (schema, metadata, knowledge base) Inaccurate AI responses on some queries — response quality and consistency need improvement Overall — audit the current implementation, discuss bottlenecks with us, and propose concrete improvements (prompting, caching, context structure, model selection, function calling, etc.) Scope of work: Audit of the current AI assistant architecture and system prompt Token consumption optimization (reducing context without losing quality) Improving the accuracy of SQL generation and natural language responses Ensuring the AI search works correctly in both the CRM and the mobile app Supporting automatic scaling — the system must work correctly with new tables and fields created by users SQL validation & security (SELECT whitelist, blocking DROP/DELETE) Proposing and implementing architectural improvements Requirements: Experience working with LLM APIs (OpenAI, Claude, Gemini, etc.) Strong SQL (PostgreSQL) skills and experience with dynamic query generation Experience with Supabase or similar BaaS platforms Node.js / JavaScript (backend) Understanding of Function Calling principles and AI-driven workflows Experience with token/context optimization for LLMs is a big plus Availability of at least 10 hours per week