Box Agent — End-to-End RAG Chatbot

An enterprise-grade Retrieval-Augmented Generation (RAG) platform that ingests content from Box, transcribes multimedia using Whisper, indexes documents with local embeddings and ChromaDB, and delivers context-aware answers through Amazon Bedrock Claude with source citations.

RAG pipeline: Box → Whisper (local) → Amazon Bedrock Claude → ChromaDB → React Chat UI

Embeddings run fully locally via sentence-transformers (no third-party embedding API needed).

Quick Start

1. Prerequisites

• Python 3.10+
• Node.js 18+
• ffmpeg installed and on PATH (required for Whisper audio)

2. Configure Environment

# Copy and fill in your credentials
cp backend/.env.example backend/.env

Edit backend/.env:

BOX_CLIENT_ID=your_client_id
BOX_CLIENT_SECRET=your_client_secret
BOX_DEVELOPER_TOKEN=your_developer_token  # Generate from Box Developer Console (valid 60 min)
BOX_ROOT_FOLDER_ID=0                      # "0" = root, or paste a specific folder ID
AWS_ACCESS_KEY_ID=your_access_key         # IAM user key or from `aws configure`
AWS_SECRET_ACCESS_KEY=your_secret_key
AWS_SESSION_TOKEN=your_session_token      # Only if using temporary STS credentials
AWS_REGION=us-east-1
BEDROCK_MODEL_ID=anthropic.claude-3-sonnet-20240229-v1:0  # Requires Bedrock model access
EMBED_MODEL=all-MiniLM-L6-v2             # local sentence-transformers model, no key needed
WHISPER_MODEL=base                       # tiny | base | small | medium

3. Install & Start Backend

cd backend
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

Backend runs at: http://localhost:8000 API docs at: http://localhost:8000/docs

4. Install & Start Frontend

# From the project root (d:\Box_agent)
npm install
npm run dev

Frontend runs at: http://localhost:3000

---

Usage

Step 1: Index Your Box Content

1. Open http://localhost:3000
2. Log in (any credentials work in this demo)
3. Switch to Developer mode (top-right toggle)
4. Click "Run Ingestion Pipeline" in the System Control panel
5. Watch progress in the Logs panel — it will fetch files, transcribe audio/video, embed chunks, and store in ChromaDB

Step 2: Chat

1. Switch back to User mode
2. Type your question in the chat bar
3. Get answers with source file citations

---

Architecture

Box API (OAuth2)
  └── box_client.py          # Recursive folder scan + download
        │
        ├── .txt / .vtt / .srt / .json  ──→  Read directly
        └── .mp3 / .mp4 / .wav / ...   ──→  transcriber.py (Whisper)
                                                    │
                                              ingestion.py
                                              (chunking + orchestration)
                                                    │
                                             embedder.py
                                          (sentence-transformers local)
                                                    │
                                           vector_store.py
                                           (ChromaDB local)
                                                    │
                                            rag_agent.py
                                      (Amazon Bedrock Claude + top-k search)
                                                    │
                                            FastAPI /api/chat
                                                    │
                                          React Frontend (ChatView)

---

API Endpoints

Method	Path	Description
GET	/api/health	Health check
POST	/api/chat	RAG query → answer + citations
POST	/api/ingest	Trigger ingestion pipeline
GET	/api/ingest/status	Live pipeline progress
GET	/api/stats	ChromaDB statistics
GET	/api/folders	Box folder tree
GET	/api/logs	Recent pipeline logs

---

Notes on Developer Tokens

Developer Tokens are short-lived access tokens valid for 60 minutes. They are ideal for quick development. For a production environment, you should transition to OAuth 2.0 with the standard authorization code grant to generate long-lived refresh tokens.

---

Whisper Model Sizes

Model	Size	Speed	Accuracy
tiny	75MB	Very fast	Lower
base	145MB	Fast	Good
small	460MB	Medium	Better
medium	1.5GB	Slow	High