📬 Personal AI Agent

Overview

This AI Agent is designed to manage your email inbox efficiently and intelligently respond to queries related to large, unstructured documents. It combines natural language understanding with retrieval-augmented generation to automate workflows and surface relevant information with minimal manual input.

This AI Agent is also capable of conducting deep research, finding the distance between two places and offering navigation between two places with the help of openrouteservice API.

This AI Agent also has the feature of autonomously browsing the web using the playwright framework and a VLM(Vision Language Model).

🚀 Features

Email Inbox Automation

• Summarizes email content and suggests quick replies from any folder/label.
• Can sort emails from any folder and move non-important emails to the spam folder.
• Given the topic the agent can send an email to any person if his email address is provided.

Document Question Answering(Offers visual reasoning based on the images provided)

• Handles large documents using chunking and vector embeddings

• Supports plain text formats, CSV files(Any other spreasheet format) and also images(.jpg,.jpeg,.png)

• Answers natural language queries with contextual precision

• Uses semantic retrieval for accurate and scalable responses

Deep Research

• With the help of this feature the user can get a complete research report, comprising of the meaning of the topic, its use cases, advantages and disadvantages.
• This feature first extracts the topic from the user's prompt.
• It then searches the web on the topic extracted
• Then all the information is converted into embedding vectors using an embedding model and the embedding vectors are stored in the vector store.
• Then we ask the LLM to generate a series of questions revolving around the topic asked by the user.
• Then we collect the answers to all these questions by utilizing the RAG Query Engine
• Then all the collected answers are compiled and the resultant is shown to the user.
• This is maintained using a dedicated MCP (Model Context Protocol) Server. (Github link: https://github.com/Anish-CodeDev/mcp_servers/tree/main/deep_research)

Integration with GPS

• With the help of openrouteservice API, the agent can calculate the distance between two points and also offers navigation between the two places
• The LLM first captures the two places entered in the user's prompt and also figures out the task requested by the user.
• The task may be either navigation or the calculation of distance between two places.
• Using the openrouteservice API, the geo-coordinates of the places are retrieved using the names of the places mentioned by the user.
• Accordingly, the geo-coordinates are fed into the API and the distance/navigation between two places is shown to the user
• This is maintained using a dedicated MCP (Model Context Protocol) Server. (Github link: https://github.com/Anish-CodeDev/mcp_servers/tree/main/maps)

Autonomous browsing

• This agent can perform various autonomous such as:
• Booking hotels(Choosing the best)
• Booking movie tickets
• Shopping for item(selects the best item based on the requirements of the user)
• Filling online forms
• Selecting specific elements
• This too is managed using a dedicated MCP(Model Context Protocol) Server. (Github link: https://github.com/Anish-CodeDev/mcp_servers/tree/main/autonomous_agent)

Unified Interface

• Both the email agent and the document(RAG) agent,maps,deep research and autonomous browsing are integrated into a single unified ReAct agent.
• The agent can be accessed by a CLI interface, by running a python file, which gives access to the ReAct agent.

🛠️ Technology Stack

Layer	Tech
Language Model	Gemini 2.5 Flash-Lite
Agentic Framework	LangGraph
Document Indexing	LlamaIndex
Vector Store	DeepLake
Email Integration	Gmail API(SMTP is used for sending emails)
Autonomous browsing	Playwright

📁 Setup Instructions

1. Clone this repo

2. Create and activate the virtual environment

3. Install dependencies: pip install -r requirements.txt

4. Fill in all the api keys where ever required.

5. Running python files: You'll have to run two python files, one is the agent.py under the email_agent folder, thr other one is server.py under the rag_agent folder, the server.py holds the instructions of the flask server. Go to the email_agent folder and run the following command: python agent.py

   Then go to the rag_agent folder and run the following command: flask --app server run