Case study
RAG Chatbot That Drafts Word Documents
A RAG chatbot that answers questions from a document corpus and helps users draft structured Word documents from retrieved content.
Situation
A corporate client produced documents that drew heavily on a large internal corpus — every draft meant hours of gathering and reconciling source material.
Task
The client wanted a chatbot that could answer questions from the corpus and turn the retrieved content into a ready-to-edit Word document, reducing drafting time.
Action
I built a RAG chatbot with an interactive document-generation layer:
- Retrieval: corpus chunked, embedded, and indexed in Qdrant; answers grounded in retrieved chunks.
- Conversational drafting: the bot interacts with the user to structure the document — outline, sections, wording — before generating.
- Word export: the final structure rendered into a formatted
.docxvia ONLYOFFICE, editable by the user.
Result
- Drafting moved from a manual gather-and-reconcile process to an interactive, assisted workflow.
- Users ended with an editable Word document, so output plugged straight into existing processes.
Confidentiality note
Client and corpus specifics stay general here.
Stack highlights
- LangChain for retrieval and the conversational flow
- Qdrant as the vector store
- ONLYOFFICE for
.docxgeneration - Python end to end