A question-and-answer assistant that answers from a body of documents. It reads a knowledge base, indexes it, and gives source-grounded answers — plus precise guidance on where to look next — wired into a client’s WordPress site.
Screenshot — pending
The Knowledge Assistant is a retrieval-augmented question-answering assistant — it answers questions from a body of documents rather than from a model’s general memory. It crawls an authenticated knowledge base, indexes the content, and provides intelligent Q&A with relevance scoring and multi-language navigation guidance, wired into a client’s WordPress site. The aim is to give end users answers that are accurate and grounded in the source material, together with precise directions on where in the underlying documentation to go next.
It is a staged pipeline of components, each handing on to the next:
A relevance-scoring and feedback subsystem self-adjusts article rankings from real user feedback, and the crawler supports deduplication, resumable runs with progress tracking, and full interaction logging.
It is written in Python. Crawling and authentication use browser automation; semantic search runs on a vector store; and answers are generated by large language models. The advanced retrieval layer is where it earns its keep: semantic chunking, hypothetical-document embeddings, automatic language detection across English, Dutch, and German, hybrid keyword-and-vector search fused by rank, and query expansion with multilingual synonyms — all toggleable from the dashboard. Relevance filtering quietly downranks archived or deprecated articles, user feedback trains the rankings over time, and the whole thing ships in containers with separate development and production configurations.