This project is a role-based chatbot application built using FastAPI, Streamlit, and LangChain + Groq LLM. It restricts users to ask questions only within their role-based access level. For example, an HR user can only query HR-related documents, while Engineering can only access Engineering-related data.

• engineering
• finance
• general
• hr
• marketing

project/
│
├── app/
│ └── services.py ← Core LLM + VectorDB logic
│
├── main.py ← FastAPI API (login, chat endpoints)
├── streamlit_app.py ← Streamlit UI (frontend)
│
├── resources/
│ └── data/ ← Department-wise document sources
│ ├── engineering/
│ ├── hr/
│ ├── finance/
│ └── marketing/
│
└── resource/
└── vectorstore/ ← Vector database for Chroma

🚀 Project Flow Overview

📁 1. Document Preparation and Processing All documents are stored in the resources/data folder, organized by department (e.g., hr, marketing, engineering).

The process() function:

• Loads all department-wise files (Markdown/CSV).
• Splits them into smaller chunks using LangChain's RecursiveCharacterTextSplitter.
• Adds metadata like department name and allowed roles.
• Stores them in Chroma vector store after generating embeddings using HuggingFace.

🧠 2. LLM and Vector Store Initialization On the first question or document processing:

• The Groq API (LLaMA 3 model) is initialized.
• Vector store is created using ChromaDB.
• Embeddings are generated using all-MiniLM-L6-v2.
• Then we can filter the documents based on the metadata , inorder to restrict the role based questions

🔐 3. FastAPI Backend (main.py) FastAPI server handles:

• User Login (/login) using basic HTTP authentication.
• Protected Chat Route (/chat):
• Checks the user's role.
• Runs a similarity search on documents using ChromaDB.
• Filters results based on user role.
• Sends allowed documents to the LLM to generate an answer.

💻 4. Streamlit Frontend (app.py) Simple login page where users can enter their username and password. On successful login:

• They are redirected to a chatbot interface and asks to enter the username and password to check the credentials , I have a dummy user database , so i restricted upto some few people
• Users can enter questions and receive role-based answers.
• Supports logout to switch between user roles.

📮 Postman Testing (Backend API) You can use Postman to test the HTTP endpoints like /login and /chat. 🔸 Screenshot: Testing /login endpoint using Postman:

🔸 Screenshot: checking wheather engineering team person can access finantial questions:

🖥️ Streamlit Frontend (Chatbot UI) Once the backend is running, you can launch the chatbot interface via Streamlit. 🔸 Screenshot: Streamlit UI after successful login

🔸 Screenshot: checking wheather engineering team person can access finantial questions:

✨ Tech Stack

• Frontend: Streamlit
• Backend: FastAPI
• LLM: LLaMA-3.3 70B via Groq API
• Vector Store: ChromaDB
• Embeddings: HuggingFace Transformers
• LangChain: Document loading, QA chain, chunking
• Postman: Manual API testing
• Dotenv: Secure API key loading

1. Clone the repository:

   https://github.com/siddu28/ds-rpc-01
   

2. Install dependencies::

    pip install -r requirements.txt
   

3. Run the fastapi server::

    uvicorn app.main:app --reload
   

4. Run the streamlit app::

    streamlit run streamlit_app.py