DSCC OpenLab · Artificial Intelligence
DocuMind — Intelligent Document Assistant
A demonstration open-source RAG application that answers grounded questions about documents uploaded by a user.
Demonstration only. This example is not a confirmed DSCC activity, publication, project, or resource. External actions remain disabled until verified information is available.
How RAG helps
RAG combines search with generation. The search step narrows a large document collection to the passages most relevant to a question. The generation step uses those passages to produce a readable response.
- Reduces reliance on unsupported model memory
- Makes document-specific questions possible
- Allows learners to inspect the evidence behind an answer
Responsible use
A production version would require file validation, access controls, retention rules, prompt-injection defenses, model monitoring, and clear warnings that generated answers can still be wrong.
Architecture
A Next.js interface uploads files and sends questions to a FastAPI service. The service extracts and chunks text, creates embeddings through a LangChain pipeline, and stores vectors in ChromaDB. At question time, the retriever selects relevant chunks and passes them with the question to a language model. The API returns an answer together with source references. This is an architectural demonstration, not a deployed DSCC system.
Main features
- Document upload and text extraction
- Semantic retrieval over document chunks
- Answers grounded in retrieved context
- Source references for checking responses
- Separation between the Next.js interface and FastAPI service
Contributors
No contributors are assigned because DocuMind is demonstration content. Verified contributor profiles would appear here if the project were approved and built.