Skip to content

Latest commit

 

History

History
57 lines (43 loc) · 2.94 KB

DOCUMENTATION.md

File metadata and controls

57 lines (43 loc) · 2.94 KB

SideQuest Documentation

SideQuest is a RAG-based document intelligence system that allows users to upload documents and query them with AI-powered answers that feature narrative tone and inline citations.

Backend Structure

The backend is built with FastAPI and handles document ingestion, vector storage, and AI-driven retrieval.

Core Modules

  • main.py: The entry point of the API, configures middleware (CORS) and includes routers.
  • requirements.txt: Lists all Python dependencies.

API Routes (backend/api/routes)

  • upload.py: Handles file uploads, background processing, and initial status tracking.
  • query.py: Orchestrates the RAG flow, from retrieval to LLM generation.
  • documents.py: Provides endpoints for listing and deleting documents.
  • init.py: Exports the routers for easy import in main.py.

Ingestion Services (backend/services/ingestion)

  • parser.py: Extracts text from PDF and DOCX files.
  • chunker.py: Splits extracted text into manageable, overlapping tokens.
  • embedder.py: Converts text chunks into mathematical vectors using Gemini.
  • database.py: Manages SQLite connection and SQLAlchemy sessions.
  • models.py: Defines the database schema for document metadata.
  • status_manager.py: Tracks the lifecycle of document processing (processing, completed, failed).

Retrieval Services (backend/services/retrieval)

  • vector_store.py: Interacts with ChromaDB for saving and searching chunks.
  • retriever.py: Performs semantic searches filtered by user_id and document_id.

Generation Services (backend/services/generation)

  • prompt_builder.py: Constructs the final narrative-style prompt for the LLM.
  • llm_client.py: Wraps the Gemini API for answering questions based on context.

Frontend Structure

The frontend is a Next.js application designed with a premium, glassmorphic UI.

Core Files

  • src/auth.ts: Configures NextAuth for session management and JWT strategy.
  • src/middleware.ts: Protects routes and ensures only authenticated users can access the dashboard.
  • src/utils/api.ts: Axios-based client for communicating with the backend API.

Components (src/components)

  • layout/Shell.tsx: The main application layout, handling the sidebar and user session.
  • views/Library.tsx: The document management view (uploading, listing, deleting).
  • views/Chat.tsx: The AI chat interface with markdown rendering and history.

Styling

  • src/app/globals.css: Global CSS variables and glassmorphism definitions.
  • src/components//*.module.css**: Scoped styles for individual components.

Cloud Deployment

The project is configured for a professional cloud environment:

  • render.yaml: Defines the backend infrastructure (FastAPI on Render.com).
  • Persistent Storage: Uses Render Disks to persist the SQLite database and ChromaDB vectors.
  • Environment Driven: All URLs and secrets are managed via environment variables on Render and Vercel.