Sharvesh Sivagnanam

ICPC Coding Platform

Competitive Programming Judge — In Development

Working as a full-stack developer on an ICPC-style platform for the Amrita ICPC organization. Auth with Keycloak, code execution via Judge0, test-case storage on MinIO, and Docker-based deployment.

Green Aura

AI-Powered E-Commerce App (Personal)

A personal multi-service e-commerce app. Built an LLM-powered chatbot with persistent chat history using LangChain, made the app fully multilingual with an LLM translation layer on the backend, and integrated Twilio for phone-number authentication. Flutter buyer app on a Spring Boot + FastAPI backend.

SoulSync

Emotion-Based Song Recommendation

A team project that recommends music matching your current mood. Detects human emotion from video and audio, extracts emotional features from songs, and uses NLP to organise tracks across languages — served through a Flutter app with a Node.js backend and a Python AI engine.

Pi-Crop-AI

Offline Edge-AI Crop Doctor — Raspberry Pi 5

Built an offline-first AI agent that diagnoses crop disease from a leaf image and generates a treatment plan — entirely without internet. Combines a TFLite CNN vision model, a locally-served LLM (Ollama), and a FAISS vector memory of past cases, with a CV-based severity estimator and a safety validator on decisions.

Music Detection

Shazam-Style Audio Fingerprinting

Built a real-time song identifier that recognises a track from a live microphone in under 5 seconds. Spectral peak constellations produce 64-bit landmark hashes; offset-alignment voting and 3-window consensus gating over a WebSocket stream keep it robust to background noise and phone-quality audio.

Treasure Hunt

Real-Time Event App

Built a real-time mobile app for treasure-hunt events, designed for 100–150 concurrent participants. Implemented GPS + QR checkpoint verification, live admin monitoring, and real-time sync with Convex.

EV Range Predictor

Machine Learning — Regression

Built a model that predicts the achievable driving range of an electric vehicle from parameters such as speed, road type, and other driving conditions — helping estimate how far a charge will realistically go.