This project showcases a Smart AI-powered WhatsApp and Mail Assistant Agent designed to automate WhatsApp message processing, meeting scheduling, and contextual chat handling using an LLM integrated through LangChain. The assistant simulates a real-time personal assistant experience with memory, task delegation, and natural conversation abilities.
It is built with Flask, LangChain, Gemini LLM, and mocked Gmail/Google Calendar APIs, showcasing how generative agents can streamline human-task interactions in daily workflows.
Sample screenshots from the assistant’s test runs:
- 🤖 Agent-based logic using LangChain & Gemini LLM
- 💬 Understands natural language WhatsApp messages
- 🧠 Maintains chat history to provide context-aware replies
- 📅 Schedules meetings, checks availability, and provides summaries
- 📨 Can classify emails (via tool) and suggest actions/reminders (optional extension)
- 🔧 Uses mocked data structures for quick prototyping and testing
| Tech | Description |
|---|---|
| Python 3.10+ | Core programming language |
| Flask | REST API handling and WhatsApp gateway |
| LangChain | LLM orchestration and tool integration |
| Gemini LLM | Response generation and reasoning engine |
| Regex | Output formatting and cleanup |
| Google Calendar API | (Optionally used via tools) |
| Gmail API | (Optional email extraction via tools) |
whatsapp_ai_agent/
├── main.py # Entry Flask app to process WhatsApp messages
├── geminillm.py # Gemini LLM wrapper for LangChain
├── agent_runner.py # Main LangChain agent setup and invocation
├── agent_config.py # Agent settings, memory, prompt loading
├── prompt_template.txt # Core agent prompt template
├── tools/
│ ├── calendar_tools.py # Calendar mock tool
│ └── email_tools.py # Email classification / summary tool
├── handlers/
│ └── google_calendar.py # API-based Google Calendar handling (extensible)
├── gmail_reader/
│ └── mail_read.py # Gmail API reader (if used)
├── utils/
│ └── helper.py # Utilities like time formatting, validators
├── static/screenshots/ # Screenshots used for the README-
Run Flask App – Accepts POST requests with user WhatsApp messages.
-
User Context History – Preserved in memory using dictionaries for accurate replies.
-
LLM Agent Chain – LangChain agent initialized with:
- Gemini LLM
- Custom tools (calendar, email)
- Structured prompt template
-
Tool Invocation – Based on user input, tools like calendar or email reader are triggered.
-
Reply Generation – The assistant replies with natural responses, using memory and tools.
The prompt_template.txt contains:
- Role Definition
- Capabilities (schedule, list, describe meetings)
- Instruction structure for using tools and memory
- Example meeting entries and reply behavior
- The assistant uses a temporary in-memory mock meeting list for all operations.
- You can test and take screenshots by modifying the mock meeting list for realistic examples.
- Real integrations can be toggled with environment configs for Gmail/GCal APIs.
- The template ensures context-switching, memory use, and tool-based logic.
This AI assistant system is part of a larger voice and WhatsApp automation solution aimed at demonstrating real-world applications of LangChain agents. It merges LLMs with task tools, stateful memory, and intuitive communication. Built to automate personal or organizational interactions with calendar/email tools in real time.