27.09.2024 - 29.09.2024Hackathon project

Rutube Support

Rutube Support

Tech stack I used:

  • React
  • NestJS
  • FastAPI
  • LLM Phi-3
  • BERT
  • Qdrant
  • Docker

AI assistant for technical support that finds the right answer in the knowledge base and drafts a ready-to-send reply on behalf of the operator

Service features

  • Fine-tuned classifier for request topic detection
  • Search for the closest answers in the knowledge base
  • Answer generation with a large language model
  • Request analytics collection
  • Average response time – 5 seconds
  • Processing over 20,000 requests per day
  • Answer classification accuracy ~85%, response accuracy ~70%

What I did on the project

  • Built frontend for request analytics
  • Created backend for statistics collection and processing

Gallery

Demo