Deep-learning and computer-vision systems, machine-learning & data pipelines, and full-stack software — plus awards and peer-reviewed publications.
Lightweight LSTM attention model for dialogue summarization on the SAMSum dataset, with a modular RL wrapper (SCST, PPO) that improved BLEU by +5.42% over baseline without pretrained transformers.
Zero-shot image classifier using CLIP image/text embeddings and a knowledge-graph R-GCN with edge-attention and a prototype-refinement loss before an MLP head. Improved over the CLIP baseline by +23% Top-1 accuracy (59% vs. 36%) and +13% F1 on AWA2.
AI-based tool for real-time exercise monitoring using BlazePose, yoga-pose classification, and automated data reporting.
Detects blood chambers in echocardiogram video frames using a U-Net segmentation approach, built during a hackathon.
AI-powered app that corrects posture in real time, tracks reps, and helps deliver safe, effective, injury-free workouts.
Forecasting tool using LinkedIn data and ML models (ARIMA, clustering) to uncover job trends and salary insights.
Document classification system built with PySpark to handle big-data volumes and optimize multi-class document processing.
Two-stage recommendation pipeline combining collaborative filtering and content-based search over 44,072 fashion products, with FAISS + CLIP embeddings for ~10ms vector retrieval and a neural re-ranker. Full-stack React 18 + FastAPI app with 11 REST endpoints; sub-100ms end-to-end latency, 95.4% recommendation diversity, 1-interaction cold-start.
Benchmarked six storage formats — CSV, Parquet, Feather, LMDB, WebDataset and TFRecord — across CIFAR-10 image training and 1M-row tabular ML workloads, quantifying impact on I/O throughput, load time, and storage efficiency while showing training accuracy stayed format-neutral.
Simple chat application using React Chat Engine with real-time messaging and authentication.
Platform integrating six independent Next.js services (fitness coaching, nutrition planning, skin analysis, restaurant discovery, fitness analytics, community) under a unified opaque-token auth system on ports 3000–3006. MongoDB-backed, Google Gemini for AI features, MediaPipe-pose exercise tracking with rep counting and form feedback. Each service ships its own Dockerfile.
Privacy-first personal-finance desktop app for Windows. Auto-categorizes transactions from bank-statement Excel exports, monitors budgets with real-time alerts, surfaces spending patterns and recurring subscriptions, and renders interactive analytics — all data stays local, no cloud sync. Built with Streamlit, Pandas, and Plotly on Python 3.11.
Full-featured e-commerce platform with customer features (browse, cart, secure Stripe checkout, order history) and an admin dashboard (sales analytics, product/order/user management). Built with Next.js 15, TypeScript, Prisma, and NextAuth.js.
ML framework for predicting tensile properties (yield strength, elongation) of miniaturized SS-316 nuclear specimens from sparse, imbalanced data. GAN + SMOGN data augmentation; Random Forest / XGBoost reached Pearson r > 0.98, establishing guidance for matching augmentation strategy to target property. Co-author, with INL collaborators.
Co-authored work demonstrating a transfer-learning deep-learning framework that outperformed four state-of-the-art models for cross-scale defect segmentation in nuclear materials.
Awarded in the AI track for "RePosture AI", a computer-vision tool for posture correction with real-time feedback.
Placed 7th of 20 teams, solving 9/13 security challenges using AWS services including Lambda, CloudTrail, DynamoDB, S3, VPC, IAM, and SageMaker.
Awarded for "Human pose estimation in fitness tracking and guidance" at VTU, Belagavi, August 2022, supported by KSCST.
Supported Prof. Mark Zhao by holding office hours, assisting with assignment/project design, grading, and helping students debug distributed-systems concepts and implementations.
Data extraction, backend development, and statistical analysis to deliver data-driven insights.
Lead the Layers Club to promote knowledge sharing, networking, and skill development in data science and AI-ML.