React
TypeScript
NestJS
Flask
QdrantLLM-powered Q&A system using knowledge base context and previously resolved questions. Responses include not only text but also supporting images and documentation links with page references for a complete answer.
Built at the Rosatom "Atomic Hack 2.0" hackathon and placed 4th
System components
- Chat website for creating requests and receiving answers
- Widget for easy embedding of AI support on any site
- Admin panel for uploading new documents (PDF) and enriching the system with examples
- Request statistics in the admin panel
Technical highlights
- Open-source SaigaLlama3 LLM running on CPU
- Text vectorization with intfloat/multilingual-e5-large
- Qdrant vector store for relevant context search
My role in the project
- NestJS backend for statistics and chat history
- API documentation for all microservices
- Backend integration with AI API for LLM