What I Built:

Local-first chess coaching app that runs as both a desktop app and a web app, importing Chess.com games into SQLite, analyzing positions with Stockfish, and using local or hosted LLMs to explain the strategic reasons behind key moments.

• Coordinated a FastAPI backend, React analysis UI and pywebview desktop shell.
• Ran a local Stockfish UCI process with deterministic chess feature extraction and background analysis jobs.
• Added local and hosted model-provider fallbacks for coaching explanations beyond raw engine lines.

Stack:

What I Built:

Implementation of an approximate LLM unlearning pipeline exploring targeted knowledge removal through reinforce, translate, relabel, and unlearn stages with GPU and CPU-friendly paths.

• Implemented the unlearning flow based on the Who's Harry Potter approach.
• Built reinforcement, anchor extraction, generic-label generation and unlearning stages.
• Supported GPU and CPU-friendly paths for experimentation.

Stack:

What I Built:

Digital-payment fraud monitoring system with UPI-style transaction simulation, behavioral risk scoring, fraud network graphs, ring detection, suspicious-node analysis, heatmaps, alerts, and explainability.

• Built a fraud monitoring workflow that scores transactions and highlights risky payment behavior.
• Detected suspicious relationships with graph analysis, fraud-ring detection and suspicious-node analysis.
• Added investigation dashboards with heatmaps, alerts and explainability for risk review.

Stack:

What I Built:

End-to-end analytics pipeline on Aadhaar enrollment data across Indian states, districts and pincodes, with Python ETL, statistical modeling, Streamlit exploration and a React dashboard.

• Built ETL for cleaning, feature engineering, state standardization, rolling averages and anomaly detection.
• Created Streamlit and React dashboards for trends, states, seasonality, districts, clusters and forecasting.
• Generated cleaned exports, publication-ready plots and a documented analysis report.

Stack:

What I Built:

Document Q&A app that indexes uploaded PDFs with tree-structured retrieval instead of traditional vector search, then answers questions from retrieved document sections with query tracing.

• Used PageIndex-style tree retrieval to reduce dependence on chunking and vector similarity.
• Built PDF indexing, retrieval traces, multi-model routing and a React/FastAPI interface.
• Connected document retrieval to grounded answer generation from retrieved sections.

Stack:

What I Built:

AI-powered API mocking platform that turns plain-English API descriptions into realistic schemas, synthetic JSON, live mock endpoints, CRUD simulation, validation, and chaos-mode failures.

• Turned plain-language endpoint descriptions into structured API schemas.
• Generated synthetic JSON, live mock endpoints, CRUD behavior and validation flows.
• Added chaos-mode failures to help frontend teams test unstable API states.

Stack:

What I Built:

End-to-end ML system for classifying cardiac risk from biomarker measurements with cross-validation, SMOTE, model comparison, SHAP explanations, FastAPI, dashboarding, and API protections.

• Processed clinical inputs and predicted heart-attack risk with model evaluation.
• Compared ML models and added explainability through SHAP outputs.
• Wrapped prediction workflows with FastAPI, dashboarding and API protections.

Stack:

What I Built:

Analytics and ML system using Bengaluru ward-level ride data to identify operational patterns, reliability, demand gaps, revenue leakage, risk classification, clustering, and SHAP explanations.

• Built feature engineering for ward-level ride data analysis.
• Added risk classification, clustering and explainability for operational patterns.
• Presented demand gaps, reliability issues and revenue leakage through a FastAPI/React interface.

Stack:

What I Built:

Decentralized bounty platform for rewarding open-source contributions with USDC on Stellar Testnet, using GitHub webhooks to release payments when qualifying pull requests are merged.

• Connected GitHub webhook events to bounty payment release logic.
• Built Stellar Testnet payment flows for open-source contribution rewards.
• Integrated wallet-oriented approval and payout behavior.

Stack: