title	Corporate RAG Bot Backend
emoji	🤖
sdk	docker
app_port	7860

🤖 Corporate-Bot — Enterprise RAG Assistant

Corporate-Bot is a production-grade, agentic Retrieval-Augmented Generation (RAG) assistant that enables users to upload corporate PDFs and interact with them using natural language.

It provides accurate, grounded answers and structured workplace actions powered by HuggingFace-hosted LLM inference.

---

🌐 Live Architecture

Frontend: Next.js (Vercel) Backend: Flask API (HuggingFace Spaces — Docker) LLM: HuggingFace Inference API (meta-llama/Llama-3.2-3B-Instruct:novita)

---

🚀 Core Capabilities

📄 Document Intelligence (RAG)

Upload any corporate PDF and:

• Ask factual questions
• Get grounded answers
• Receive structured responses
• Prevent hallucinations

Supports:

• Text
• Tables
• Structured content

---

⚙️ Agent-Based Reasoning

The system intelligently detects user intent:

Example:

User Input

Create a ticket for VPN not working

Output

{
  "action": "create_ticket",
  "department": "IT",
  "priority": "High",
  "description": "VPN not working"
}

---

🛡️ Hallucination Control

Strict refusal logic:

If answer not present:

Information not found in uploaded document.

No guessing. No fabricated answers.

---

🧠 Architecture Overview

PDF Upload
   ↓
Unstructured Parser
   ↓
Structure-Aware Chunking
   ↓
Embedding (BGE)
   ↓
FAISS Vector Store
   ↓
Retriever
   ↓
Cross-Encoder Reranker
   ↓
Agent Supervisor
   ↓
HuggingFace LLM Inference
   ↓
Final Response

---

🧰 Technology Stack

Component	Tool
Backend	Flask
Frontend	Next.js
Embeddings	BAAI/bge-small-en
Vector DB	FAISS
Reranker	cross-encoder/ms-marco
LLM	HuggingFace Inference
PDF Parser	Unstructured
Hosting	HuggingFace Spaces
Frontend Hosting	Vercel

---

🌐 Live API Endpoints

Health Check

GET /

Response:

{
 "success": true
}

---

Upload PDF

POST /api/v1/upload

---

Ask Question

POST /api/v1/chat

---

🌍 Live Deployment

Backend:

https://AyushAnand413-corporate-rag-bot-backend.hf.space

Frontend:

(Your Vercel URL)

---

🧪 Example Queries

Factual

What is the vision of 6G networks?

---

Table-based

What was revenue growth in FY25?

---

Conceptual

What are key risks mentioned?

---

Action

Create a ticket for VPN not working

---

🔐 Security & Safety

✔ No hallucinated data ✔ Evidence-based answers ✔ Strict refusal logic ✔ Secure inference via HF Token

---

🧑‍💻 Local Development

Requirements

Python 3.10+

---

Install

pip install -r requirements.txt

---

Run

python web_app.py

Open:

http://localhost:7860

---

🔑 Environment Variables

Required:

HF_TOKEN=your_token

Optional:

HF_GENERATION_MODEL=meta-llama/Llama-3.2-3B-Instruct:novita
ALLOWED_ORIGINS=*

---

📁 Project Structure

Corporate-bot/
│
├ agent/
├ ingestion/
├ retrieval/
├ llm/
├ frontend/
├ web_app.py
├ Dockerfile
└ requirements.txt

---

📈 Production Features

✔ Docker deployment ✔ CI/CD via GitHub Actions ✔ HF Spaces hosting ✔ Vercel frontend ✔ Runtime PDF ingestion ✔ API-first backend

---

🎯 Use Cases

Enterprise assistants Corporate document search Legal document QA Financial report analysis Internal automation bots

---

🧠 Model Used

meta-llama/Llama-3.2-3B-Instruct:novita

Hosted via:

HuggingFace Inference API

---

👨‍💻 Author

Ayush Kumar Anand

Swarnim Vatsyayan

Shuchay Viswanathan

---

⭐ Conclusion

Corporate-Bot is a fully production-ready enterprise AI assistant combining:

• Retrieval-Augmented Generation
• Agent-based reasoning
• Secure cloud inference
• Modern frontend architecture

---