FDE Helper Bot

A JSON-based RAG chatbot that helps team members instantly find the right person to contact for any issue. Built with FastAPI, BM25 retrieval, and Portkey AI — themed with the Portkey Design System.

How It Works

User Query
    │
    ▼
┌──────────────┐     ┌───────────────────┐     ┌────────────────────┐
│  BM25 Index  │────▶│  Retrieve top-k   │────▶│  Inject context    │
│  (built from │     │  matching team     │     │  into system       │
│  JSON KB)    │     │  entries by score  │     │  prompt            │
└──────────────┘     └───────────────────┘     └────────┬───────────┘
                                                        │
                                                        ▼
                                               ┌────────────────────┐
                                               │  Portkey Chat      │
                                               │  Completion (SSE   │
                                               │  streaming)        │
                                               └────────┬───────────┘
                                                        │
                                                        ▼
                                                  Streamed Response

1. Knowledge Base — A flat JSON file (knowledge_base.json) stores team members with their name, role, and a natural-language description of what they handle.
2. BM25 Retrieval — At startup, descriptions are tokenized into a BM25Okapi index. Each user query is scored against all entries and the top matches are selected as context.
3. LLM Generation — The retrieved context is injected into a system prompt and sent to an LLM via Portkey, which streams a response back to the UI over SSE.

Team Directory

Name	Role	Handles
Visarg	MCP Specialist	MCP server config, integration debugging, protocol, tool registry
Avanish	Solution Architect	Deployments, infra, CI/CD, Docker, K8s, DNS, SSL, DevOps
Ayush	CTO	Engineering escalations, critical decisions, production incidents

Project Structure

FDE_helper/
├── app.py                     # FastAPI backend (BM25 retrieval + Portkey streaming)
├── knowledge_base.json        # Team knowledge base (flat JSON)
├── portkey-theme.css          # Portkey Design System reference
├── templates/
│   └── index.html             # Chat UI (Portkey-themed, vanilla HTML/CSS/JS)
├── assets/
│   ├── chat_home.png          # Screenshot — home screen
│   └── chat_conversation.png  # Screenshot — active conversation
├── requirements.txt           # Python dependencies
├── .env                       # Environment config (not committed)
└── README.md

Setup

Prerequisites

• Python 3.10+
• A Portkey API key with a configured virtual key

Installation

git clone git@github.com:harshitkumar454/FDE-Helper.git
cd FDE_helper

python -m venv venv
source venv/bin/activate

pip install -r requirements.txt

Configuration

Create a .env file in the project root:

PORTKEY_API_KEY=your_portkey_api_key_here
MODEL=@your-virtual-key/provider/model-name

Running the App

python app.py

Open http://localhost:8080 in your browser.

API Endpoints

Method	Path	Description
GET	/	Serves the chat UI
POST	/chat	Streaming chat endpoint (SSE)
GET	/team	Returns the knowledge base as JSON

POST /chat

Request:

{
  "messages": [
    { "role": "user", "content": "MCP server is down, who do I contact?" }
  ]
}

Response (Server-Sent Events):

data: {"content": "You"}
data: {"content": " should"}
data: {"content": " contact"}
data: {"content": " Visarg..."}
data: [DONE]

Customizing the Knowledge Base

Edit knowledge_base.json to add or update team members:

{
  "name": "Person Name",
  "role": "Their Role",
  "description": "Natural language description of what they handle. BM25 matches against this text."
}

Restart the server after editing — the BM25 index rebuilds automatically.

Tech Stack

Component	Technology
Backend	FastAPI + Uvicorn
Retrieval	BM25Okapi (rank-bm25)
LLM Gateway	Portkey AI
LLM Model	Llama 3.3 70B (via OpenRouter)
Frontend	Vanilla HTML/CSS/JS + SSE
Theme	Portkey Design System

License

MIT