Video has become the way organizations record what they know, and the least searchable thing they own. This Specialization covers Video RAG: making that content retrievable and answerable.
You extend retrieval augmented generation from documents to video, breaking recordings into frames, audio, transcripts, captions, and on-screen text, then turning those into searchable records. You retrieve precise timestamped moments and connect them to a model that answers questions and cites its source.
By the end of this Specialization, you will be able to:
1. Extract frames, audio, transcripts, captions, and OCR text from video.
2. Generate multimodal embeddings and index them in a vector store.
3. Implement semantic, metadata, and hybrid retrieval across a collection.
4. Build grounded question answering and a chat assistant.
5. Measure quality with precision, recall, and groundedness.
6. Deploy the pipeline through a FastAPI backend.
This Specialization suits AI engineers, machine learning engineers, data engineers, and backend developers who already work with language models and want to extend that work to video, along with technical teams sitting on large video archives. It assumes working Python and comfort with notebooks, and no background in computer vision, speech, or vector databases.
Enroll now to build a Video RAG system that answers questions about your own video library.
Projet d'apprentissage appliqué
Across the Specialization, you complete hands-on projects that mirror the work of building a video retrieval system end to end. You process raw video into frames, audio, transcripts, captions, and on-screen text, then generate embeddings and index them in a vector database. You build search across a video collection, connect retrieval to a language model so that answers stay grounded and carry timestamps, and extend it into an assistant that handles follow-up questions. In the final project you evaluate retrieval and answer quality against a dataset you build yourself, apply re-ranking and latency optimizations, and serve the finished pipeline through a FastAPI backend and web interface. Each project produces a concrete artifact, including pipelines, indexes, evaluation reports, and a running application, that you can add to a professional portfolio as evidence of practical skill.
