Projects
Selected Work
A collection of my major works — blending research, AI innovation, and production deployments.
Fine-tuned EfficientNetB0 on Food-101 (101K images) achieving 82.57% test accuracy. Health-profile engine for 5 dietary types with structured JSON responses. FastAPI backend integrates USDA FoodData Central (300K+ foods) + 30+ Indian dishes. Full-stack deployed on Vercel + Render.
NLP pipeline extracting sentiment from Reddit using NLTK + VADER. Combined with Yahoo Finance historical data to train XGBoost classifier — improved prediction accuracy from 58% to 63%. Streamlit dashboard for real-time visualization of sentiment trends.
A minimal MLOps-style batch job demonstrating reproducibility, observability, and deployment readiness. Loads config from YAML, reads 10K-row OHLCV CSV, computes rolling mean on close price, generates binary trading signals, writes structured metrics.json + detailed run.log. Runs locally and inside Docker.