Company Onboarding Agent

This project is a complete AI-powered Employee Onboarding Agent using LangGraph, FastAPI, React, Telegram, MongoDB, and Supabase (pgvector).

Features

• Telegram Bot: New employees interact with an AI bot to ask questions and track their onboarding progress.
• HR Dashboard: HR can view all employees, monitor chats, see task progress, and upload documents (PDFs) to the Vector Knowledge Base.
• AI Router: LangGraph routes the user's intent to Q&A (RAG), Progress Tracker, or Smalltalk.

Tech Stack

• Backend: FastAPI, LangChain, LangGraph
• LLM: Google Vertex AI (Gemini 3.1 Pro)
• Vector DB: Supabase + pgvector
• Database: MongoDB (Users, Chats)
• Frontend: React, Vite, Tailwind CSS

Getting Started

1. Backend Setup

1. cd backend
2. Create virtual environment: python3 -m venv venv
3. Activate it: source venv/bin/activate
4. Install dependencies: pip install -r requirements.txt
5. Configure .env:

   PORT=8000
   TELEGRAM_BOT_TOKEN=your_bot_token
   MONGO_URI=mongodb://localhost:27017
   MONGO_DB_NAME=onboarding_db
   SUPABASE_URL=your_supabase_url
   SUPABASE_KEY=your_supabase_key
   SUPABASE_DB_URL=postgresql://postgres:password@db.supabase.co:5432/postgres
   VERTEX_PROJECT_ID=your_gcp_project_id
   VERTEX_LOCATION=us-central1

6. Run Backend: uvicorn app.main:app --reload
7. Set Telegram Webhook to your Ngrok/Server URL: curl -X POST https://api.telegram.org/bot<TOKEN>/setWebhook -d "url=https://<NGROK_URL>/telegram/webhook"

2. Frontend Setup

1. cd frontend
2. Install dependencies: npm install
3. Run Development Server: npm run dev

Open http://localhost:5173 to view the HR Dashboard.

New Features Added

1. HR-Initiated Invite Workflow

• HR can now generate a unique invite link from the dashboard
• Click the + button in the "New Hires" sidebar
• Enter the new hire's name and click "Generate Invite Link"
• Share the generated t.me/DayOne11bot?start=XXXX link with the employee
• When they click the link and start the bot, they get a personalized DayOne welcome message

2. Merged Knowledge Base

• All uploaded PDFs are now consolidated into backend/knowledge_base_merged.txt
• This file serves as a master reference for the search tool

3. Agent Search Tool (search_merged_file)

• The agent now has access to a tool that can search the merged knowledge base
• Returns complete paragraphs (up to 20 results) containing the keyword
• Automatically triggered by the LLM when vector search is insufficient

4. DayOne Identity

• The bot now consistently identifies as "DayOne"
• All prompts updated to reflect the onboarding assistant persona