Projects, in depth

Visum

A driver monitoring system built for vehicles that have none. A camera watches the driver, not the road, and escalates from a gentle warning to an SOS.

What it does

• Tracks eye closure, head pose, gaze and yawning in real time to spot fatigue and distraction.
• Escalates in stages: a soft warning, then a stronger alert, then an SOS that shares live location.
• Logs every event with what happened, when and where.

How I built it

• Real-time vision pipeline in Python with OpenCV, MediaPipe and TensorFlow Lite, using EAR-based fatigue detection and facial landmarks.
• IoT fleet platform on React, Firebase, Node.js and ESP32 for live vehicle monitoring and alerts.
• Cloud-hosted fleet dashboard with secure sign-in and central incident management. Tested on an actual bus.

FoodGuard

A food safety analytics dashboard that turns contamination records across Indian states into filters, charts and plain-language insights.

What it does

• State-wise, year-wise and category-wise views of food contamination cases from 2018 to 2024.
• An insights panel with a summary, anomaly detection, a next-year forecast and recommendations.
• A data explorer for filtered records.

How I built it

• React and Recharts on the front, Node.js and Express REST APIs behind it, MongoDB for the data.
• Multi-level filtering in the API, so every chart is computed from live data.
• A statistical insights engine using anomaly detection and linear regression.

LedgerSync

An AI finance controller: a reconciliation agent that cross-checks a bank statement, a Razorpay settlement report and the internal ledger, then shows its work on a live dashboard.

What it does

• Matches transactions across three sources that a finance team normally checks by hand.
• Reports its own match rate, amount reconciled, exceptions, and precision and recall against a hidden ground truth.
• Regenerates its dataset with injected edge cases on every run, so results are not cherry-picked.

How I built it

• A tiered matching cascade: deterministic rules first, and an LLM only for the genuinely ambiguous cases.
• Multi-provider AI support (Claude, Gemini, Groq) behind one interface.
• Python and FastAPI backend with tests, and a React and Vite dashboard.

PlacementPilot AI

A placement assistant that reads recruitment emails for you, pings you on Discord, and applies on its own if you keep missing the reminders.

What it does

• Monitors Gmail for placement emails and extracts company, role and deadline with an LLM.
• Sends instant alerts through the Discord API.
• After 3 missed reminders, an autonomous fallback fills in and submits the application.

How I built it

• Generative AI and LLM APIs for structured extraction.
• Browser automation with event-driven scheduling for the fallback flow.
• Deployed as Docker containers.

Social Connect

An AI-based social media platform built during my internship at Arcoiris Logics.

What I built

• Real-time interactions on React, Node.js, Express and MongoDB.
• Google Gemini API integration for AI image generation.
• RESTful APIs and MongoDB data models behind the app.

PacketX

A deep packet inspection engine with a live analytics dashboard.

What it does

• Multi-threaded packet processing.
• Flow tracking using 5-tuple hashing.
• TLS SNI extraction and rule-based packet filtering.
• A live analytics dashboard in React, backed by MongoDB.

The Third Eye

AI smart glasses that help people with low vision understand what is in front of them.

What it does

• Object detection and real-time scene understanding.
• Voice guidance that describes the scene.
• Built on TensorFlow and OpenCV, running on a Raspberry Pi.

Netra

A smart traffic control management system that reads live traffic at a junction and adapts the signals to keep it moving. It won us first place at GeeksforGeeks hackathons.

What it does

• Estimates vehicle density on each lane from camera feeds.
• Adjusts green-light timing to clear the busiest lanes first.
• Gives priority to emergency vehicles.

How I built it

• Vehicle detection with OpenCV.
• Signal control logic with a small dashboard to watch each junction.

IoT Fleet Safety Tracker

ESP32 sensors on each vehicle feed a dashboard with real-time monitoring, instant alerts and status logs, so a fleet manager sees problems as they happen.

What it does

• Streams location, speed and sensor readings from every vehicle to one dashboard.
• Sends instant alerts for harsh braking, overspeeding and crashes.
• Keeps a status log per vehicle for audits.

How I built it

• ESP32 boards with GPS and motion sensors, publishing over Wi-Fi.
• Firebase for real-time data, and a React dashboard for monitoring.

Pothole Detection Rover

A small IoT rover that drives a stretch of road, detects potholes, and reports each one with its location so it can be fixed.

What it does

• Scans the road surface with ultrasonic sensors and a camera.
• Flags potholes and tags each with GPS coordinates.
• Sends reports to a web map for the maintenance team.

How I built it

• Arduino-based rover chassis with an ESP32 for connectivity.
• Simple computer vision in Python to confirm detections.