NezT Labs: Multi-Agent AI Assistant

NezT Labs (in-house R&D) · AI Workflow Software

An in-house build: a production-oriented multi-agent RAG system with five specialised roles (Router, Researcher, Planner, Coder, Responder).

What it does

FastAPI backend with synchronous and streaming chat, session management, document upload and knowledge-base listing.

Three cache tiers on Redis Stack (research, vector, response) with automatic invalidation.

PostgreSQL conversation history, context-aware windowing, multimodal input and per-model token and cost analytics.

Live streaming of agent thoughts so every step is inspectable.

Tech stack

Python, LangGraph, FastAPI, Server-Sent Events, PostgreSQL, Redis Stack, ChromaDB, OpenAI / Gemini, Docker, GitHub Actions

Impact

Fewer repeated research and model calls through multi-tier caching, lowering latency and API spend. Ships as a runnable, containerised system rather than a notebook prototype.