Skip to content

Knowledge Seeker

Project start: 2025-04-16

Project description

Knowledge Seeker is an advanced tool for transcription, indexing, and information retrieval from video recordings. As the project leader, I coordinate the development of a system utilizing the latest AI technologies for speech-to-text processing and implementation of advanced semantic search mechanisms. The application enables users not only to find specific information in extensive video resources but also to generate responses to queries based on accumulated knowledge using the RAG (Retrieval-Augmented Generation) architecture.

Preliminary architecture logic

alt text

Main functionalities

  • Transcription of video recordings to text with preservation of time metadata
  • Processing transcriptions through chunking and generating embeddings
  • Vector database for storing and efficiently searching embeddings
  • User interface enabling both simple and semantic content searching
  • RAG (Retrieval-Augmented Generation) system for generating responses to user queries
  • Deployment in Digital Ocean cloud ensuring scalability and availability
  • Data export in JSON formats and streaming capability to user API

Development Roadmap

  • Integration with additional data sources (documents, presentations, audio)
  • Enhancement of RAG mechanisms with advanced filtering and re-ranking techniques
  • Implementation of components for automatic verification and updating of the knowledge base
  • Optimization of indexing and search processes for larger datasets
  • Development of API interface enabling integration with external applications

Skills

  • Python
  • Docker
  • Digital Ocean
  • LLM (Large Language Models)
  • Natural Language Processing
  • Vector Databases
  • RAG (Retrieval-Augmented Generation)
  • REST API
  • Streamlit
  • JSON/Embeddings
  • Whisper (Speech-to-Text)
  • PostgreSQL
  • Microservice Architecture
  • Qdrant/Weaviate

Technologies used in the project

  • OpenAI API
  • Whisper for audio transcription
  • Qdrant/Weaviate as vector database
  • LangChain for RAG implementation
  • FastAPI for backend services
  • Streamlit for user interface
  • Docker for containerization
  • Digital Ocean for hosting