CS undergrad at MSU Baroda with a parallel BS in Data Science from IIT Madras. Building end-to-end software, from React interfaces and REST APIs to the machine-learning models behind them, with most of the recent work in deep learning.
- Deep learning: PyTorch and torchvision; CNNs, U-Net and GAN architectures for computer vision
- Machine learning: Scikit-Learn,Numpy,Pandas,XGBoost, LightGBM and CatBoost; supervised learning, anomaly detection, feature engineering and model evaluation
- Full-stack: React + TypeScript frontends, Node.js / Express and FastAPI backends, MongoDB and PostgreSQL, JWT authentication
- ML in production apps: models served as FastAPI microservices and consumed by Node.js backends over REST
| Project | What it does | Result |
|---|---|---|
| Chest X-ray Disease Classification | 20-class thoracic pathology classification on 51K+ X-rays. Fine-tuned DenseNet-201 with focal loss and per-class decision thresholds tuned for a metric that heavily penalises missed diseases. | Test score improved from −5.22 → −5.04 through threshold tuning |
| Mechanical Parts Segmentation | 7-class semantic segmentation of industrial images. U-Net++ with ResNet-34 and EfficientNet-B1 encoders, trained from scratch, ensembled with test-time augmentation. | Dice 0.9796 on the test set |
| Plant Leaf Super-Resolution | 4× SRGAN (32 → 128 px). SRResNet generator and conditional discriminator, with Charbonnier, Sobel-gradient, perceptual and delayed adversarial losses. | MAE 16.77 on 0–255 pixels |
| Heavy Equipment Price Prediction | Price regression on 138K sparse, high-cardinality auction records. Stacked XGBoost, LightGBM and CatBoost with leak-free in-fold target encoding. 13 versions over 3 months. | RMSLE 0.1928 on the test set |
| Project | What it does |
|---|---|
| AI Transaction Analytics Platform | React + TypeScript dashboard and an Express/MongoDB API with JWT auth. A separate FastAPI service runs anomaly detection (Isolation Forest), spend forecasting, and TF-IDF expense categorisation. |
| AI Mental Health Support Platform | Journaling, self-assessments and an OpenAI-powered chat, built on React, Express and MongoDB with role-based access control. |
Languages
Machine learning & deep learning
Web & backend
Databases & tools
| Title | Details |
|---|---|
| Second Runner-Up (3rd place) Adani Innovation Mindstorm 2026 |
Built the ML pipeline (Isolation Forest + LSTM) for anomaly detection in industrial control systems |
| Top 40 Finalist Odoo × MSU FootPrints'25 Hackathon |
|
| Open Source Contributor GirlScript Summer of Code 2026 |
Issues, features and PRs through reviewed GitHub workflows |
| Junior Training & Placement Coordinator FTE, MSU Baroda |
Coordinating recruitment drives between students and recruiters |