RAG-Based-AI-Teaching-Assistant

The RAG-Based AI Teaching Assistant is a local Retrieval-Augmented Generation (RAG) system that converts course videos into searchable knowledge and allows students to ask questions directly from the processed course content. Instead of only answering a question, the assistant also identifies which lecture contains the topic and at what timestamp it is explained. The current version is a local course-video RAG pipeline that uses DeepSeek R1 via Ollama for generation, BGE-M3 for embeddings, and OpenAI Whisper for speech-to-text. The current Whisper configuration translates Hindi audio into English transcripts. The project processes local course videos by extracting MP3 audio using FFmpeg, converting lecture audio to text using Whisper, preserving lecture timestamps, merging small Whisper segments into larger context-rich chunks, generating embeddings using BGE-M3 through Ollama, retrieving relevant course content using cosine similarity, expanding retrieval with nearby chunks from the same lecture, and generating final answers using DeepSeek R1 through Ollama. It uses only the retrieved course context as the source of truth and returns lecture title, lecture number, and timestamp. The system runs locally after setup using Ollama, and course context is not sent to an external LLM API in the current local configuration.