Loading...
© 2026 All rights reserved.
A full-stack web application that uses large language models to automate email processing through visual workflows. Users connect Gmail or Microsoft 365 accounts, build node-based automation flows, and let AI classify incoming emails, generate contextual responses, and execute scheduled actions.
fullstack
Internal / Client
6 tools
Featured Project
This project was developed as my end-of-studies internship (PFE) at Proxym, where the goal was to explore how large language models could be applied to practical business automation scenarios.
Organizations handling high email volumes lacked a flexible automation tool that could understand message context, classify content accurately, and take appropriate action without manual intervention for routine emails.
Sole full-stack developer — designed the system architecture, implemented the FastAPI backend with modular service layers, built the React frontend with the visual flow editor, integrated LangChain with Ollama for AI operations, and set up OAuth 2.0 flows for Gmail and Microsoft 365.
Built a visual workflow automation platform where users design AI-powered email processing pipelines using a drag-and-drop node editor. The system integrates Ollama LLM via LangChain for classification and response generation, connects to Gmail and Microsoft 365 via OAuth 2.0, and executes workflows on a scheduled cron basis.
Automated email processing with AI classification into user-defined categories, AI-generated contextual replies, and scheduled execution of multi-step workflows across connected email accounts.
The platform provides a low-code environment where users design email automation workflows using a drag-and-drop node editor powered by React Flow. Workflow nodes include AI-powered classifiers that sort emails into user-defined categories, AI response generators that draft contextual replies, conditional branches, and action nodes that send emails via connected provider APIs. The backend runs on FastAPI with PostgreSQL, using LangChain to interface with a locally-hosted Ollama LLM (Llama 3.2) for all AI operations. A cron scheduler executes active workflows every 30 seconds, fetching new emails from connected accounts and running them through the workflow pipeline. Real-time WebSocket updates keep the frontend synchronized with backend state changes.
Proxym
Feb 2025 — Jun 2025
A data analysis and machine learning project exploring the World Happiness Report dataset. Uses linear regression, polynomial regression, and K-Means clustering to identify factors influencing global happiness scores and group countries by well-being patterns.
A modern web-based room reservation and management system built with FastAPI and React. Allows organizations to manage bookable spaces, browse available rooms, make reservations with interactive calendar views, and provides admin tools for managing rooms, equipment, users, and reservations.