Portfolio & Credentials

Gundata: Agentic AI

Fall 2025 · Reinforcement Learning Testbed

A traditional Andhra Pradesh Sankranti dice game rebuilt as a testbed for AI agents — evaluating rule-based bots against a trained reinforcement learning opponent with live explainable decision-making. Demonstrates core ML engineering: clean MDP formulation, training at scale, and interpretability for downstream analysis.

• Tabular Q-learning agent trained over 20,000 self-play episodes on an 18-state MDP with convergence analysis
• Explainability dashboard logging each agent's reasoning and top-3 Q-values per move in real time
• Clean MVC game engine with two rule variants and multiple AI opponent personalities
• Full technical report and reproducible training pipeline for agent evaluation

MediTrust — Explainable AI Healthcare Risk Prediction

Spring 2026

Explainable cardiovascular risk prediction using Logistic Regression — ROC-AUC 0.87 with per-patient SHAP explanations for clinical interpretability. Full-stack deployment: FastAPI backend, React frontend, PostgreSQL database, AWS hosting, and GitHub Actions CI/CD for reproducible model serving.

• Logistic Regression model achieving ROC-AUC 0.87 on a held-out test set
• Per-patient SHAP explanations so every prediction includes a feature-contribution breakdown for clinical trust
• Full-stack production system: FastAPI backend, React UI, PostgreSQL persistence, Dockerized services
• AWS deployment with GitHub Actions CI/CD for automated testing and model versioning

LLM Evaluation — Handshake AI

Apr 2026–Present

Comparative evaluation of frontier LLMs in an agentic coding environment — custom task design, structured scoring, and detailed qualitative rationales that feed model improvement.

Campus-Objects — Multi-Class Object Detection

Fall 2025

LW-DETR multi-class detection on 3,000+ annotated campus images — 0.71 mAP@0.5, COCO-format pipeline with transfer learning and augmentation.

• Built and annotated a 3,000+ image campus dataset in COCO format with consistent labeling guidelines
• Fine-tuned LW-DETR via transfer learning, reaching 0.71 mAP@0.5 across multiple object classes
• Applied augmentation and evaluation pipelines for reproducible detection benchmarking

GeoAI Flood Detection System

Fall 2025

Semantic segmentation for flood-affected zones using attention U-Net on Sentinel-1 SAR imagery, with Grad-CAM visualizations for interpretability.

• Attention U-Net semantic segmentation on Sentinel-1 SAR imagery for flood-zone mapping
• Grad-CAM activation maps to verify the model attends to true flood signatures, not artifacts
• Reproducible PyTorch training pipeline built during AI-CHESS Lab research

Snap Tune — AI Music Recommender

Spring 2025

Multi-modal music recommender — BLIP image captioning + GPT mood inference + Spotify retrieval, deployed on Streamlit Cloud.

• BLIP image captioning for semantic scene understanding from user-uploaded photos
• DistilGPT2 mood inference bridging vision output to music-relevant emotional context
• Spotify API retrieval and playlist generation, deployed as an interactive Streamlit Cloud app

Gundata: Sankranti Dice

Interactive systems

Interactive recreation of a traditional Andhra Pradesh Sankranti dice game — turn-based scoring, desktop UI, and game-logic design alongside ML work.

• Turn-based dice gameplay with traditional scoring rules
• Java Swing desktop interface and state management
• Evidence of interactive frontend + game-logic skills