The problem
Teams wanted one assistant that could answer from very different sources — PDFs, company web pages and recorded videos — and let non-developers decide what it knows.
What we built
- An admin panel where knowledge sources are added as files, links or YouTube URLs.
- An ingestion pipeline that extracts text, splits it into passages and stores embeddings in a vector database.
- A chat interface that answers with retrieved passages and keeps conversational memory.
How it works
Each source is processed once and indexed. At question time, the most relevant passages are retrieved and passed to the language model with instructions to answer only from them.