AI Document Assistant
RAG document assistant that ingests business documents and generates grounded answers with citations linked to retrieved source content.
01. Overview
RAG document assistant that ingests business documents and generates grounded answers with citations linked to retrieved source content.
02. Interface / Screenshots
No dedicated project screenshot is currently in the repository; this preview uses the shared visual placeholder.
03. Architecture
Interactive Architecture Walkthrough — a visualization of the implemented architecture/workflow.
Interactive Architecture Walkthrough
A document path from upload and ingestion through retrieval, grounded generation and citation-backed streaming. [Visualization of the implemented document workflow.]
04. Problem
Teams need grounded answers from internal documents without manually searching through source files.
05. Solution
Built a Next.js and FastAPI RAG chatbot that processes uploaded documents, stores embeddings in Chroma, and returns citation-backed answers tied to source passages.
Backend focus
- -FastAPI service with PostgreSQL persistence for organizations, users, documents, chats, and messages
- -Background ingestion pipeline for extraction, chunking, embedding, and vector upsert
- -Streaming chat events for tokens, citations, and completion over SSE
- -12 backend tests recorded in the repository verification log across auth, tenant isolation, retrieval, grounding, uploads, and rate limits
AI focus
- -History-aware retrieval query construction
- -OpenAI-compatible embeddings and optional chat generation
- -Grounded answer generation from Chroma matches
- -Traceable citations with source document, page, snippet, and score
06. Key Features
- >PDF, DOCX, TXT, and Markdown ingestion
- >JWT access and refresh authentication
- >Organization-scoped documents and chats
- >Background extraction, chunking, and indexing
- >Embedding generation and storage in Chroma
- >Citation-backed SSE responses
- >Chat history and resumable conversations
- >Document, folder, tag, and date filters
- >Grounding checks and explicit fallback answers
- >Request IDs, rate limits, and upload validation
07. Engineering Challenges
Added retrieval filters and explicit grounding checks before response generation.
Returned citation-backed snippets with the answer payload.
08. Outcome
Created a document-question-answering workflow where responses can be traced back to retrieved source passages rather than presented as unsupported model output.