Context
neuro_apk is an academic full-stack prototype of an AI employee for agribusiness. The project connects a frontend, backend, crawler, database, vector search, and local model runtime.
Problem
Agribusiness knowledge work can involve fragmented sources, repeated questions, and domain-specific documents. The prototype explores how a RAG system can make that information more usable.
Constraints
- Local LLM integration through LM Studio.
- Full-stack architecture instead of a single notebook.
- Reproducible service setup with Docker Compose.
- RAG pipeline backed by a vector database.
Approach
The system uses a Next.js frontend for interaction, a FastAPI backend for orchestration, PostgreSQL for structured data, Qdrant for vector retrieval, and a crawler for collecting source content.
Implementation highlights
- Separation between UI, API, storage, retrieval, and inference.
- Docker Compose setup for a reproducible local environment.
- Vector search added as a first-class system component.
- Backend designed around API boundaries rather than notebook-only logic.
Result
The project is a working prototype that demonstrates end-to-end AI system architecture from frontend to vector search and local inference.
What it demonstrates
RAG architecture, local LLM integration, Docker-based delivery, and full-stack AI MVP thinking.