Virtual AI Manager (VAM)

Developed an autonomous multi-agent system (VAM) that functions as a "Virtual Manager" to orchestrate complex workflows, integrating 3+ specialized agents (Task Execution, People Ops, Risk Assessment) using a Directed Acyclic Graph (DAG) architecture. Implemented advanced NLP capabilities using Retrieval-Augmented Generation (RAG) to interpret user commands and contextually retrieve past decisions, enhancing response relevance and continuity across sessions. Engineered a persistent memory layer using PostgreSQL and pgvector to store and retrieve vector embeddings of user interactions, enabling the agent to maintain long-term context and improve decision quality over time. Built a robust "Human-in-the-Loop" governance system with a dedicated Risk Agent and approval workflow, preventing unauthorized high-stakes actions (e.g., deleting repositories) by enforcing manager review via a React-based intervention UI. Designed a modular backend architecture with FastAPI, seamlessly integrating external tools like GitHub (issues/PRs), Slack (bi-directional chat), and Google Calendar (schedule management) through secure OAuth2 flows.

Intelligent Traffic & VANET System

Developed a real-time object detection pipeline using the YOLO model and OpenCV to identify vehicles and pedestrians, achieving high inference speeds for traffic analysis. Engineered safety features including Lane Detection algorithms and Forward Collision Warnings, processing video feeds to calculate distances and trigger alerts. Simulated a VANET (Vehicular Ad-hoc Network) environment to model vehicle-to-infrastructure communication, optimizing traffic control systems and automated signal timing.

AI_Agent_cal

This project is a Flask-based API that integrates OpenAI's function calling with Google Calendar. The application provides endpoints to create, read, update, and delete calendar events. It uses OpenAI's language model to process event-related requests and validate actions before interacting with Google Calendar through the API.