projects.

built at ummahhacks. a full-stack MVP connecting donors directly to verified recipients. supports mobile wallets, bank transfers, cash-out vouchers, and USSD claim codes. role-based dashboards for donors, recipients, verifiers, and admins with supabase auth, react hook form + zod validation, and recharts analytics.

a personal full-stack project that lets donors send money directly to verified recipients. features mobile wallets, bank transfers, cash-out vouchers, USSD claim codes, and role-based dashboards for donors, recipients, verifiers, and admins. built with supabase auth, react hook form + zod validation, and recharts analytics.

full-stack resume analysis platform using react + node.js/express. implemented an NLP scoring engine with TF-IDF and cosine similarity evaluating skills match (40%), experience relevance (30%), keyword overlap (20%), and ATS friendliness (10%).

a multi-agent AI system that turns vague goals into actionable task plans. coordinates three specialized LLM agents (planner, critic, improver) running sequentially via supabase edge functions.

built with python and streamlit, supporting multi-horizon analysis (1m to 1y), key price metrics, interactive plotly charts with 20/50-day moving average overlays, and light/dark mode.

a lightweight python CLI tool for fetching real-time weather data with clean terminal output. features refactored weather fetching logic and structured data display.

an ensemble ML trading system combining logistic regression, random forest, and gradient boosting to rank 50 stocks monthly across 10 engineered features. achieved +67.7% cumulative return vs +46.3% S&P 500 over 23 months using walk-forward validation, kelly criterion sizing, and macro regime detection.

ML match prediction pipeline using logistic regression trained on 10+ years of T20I data across 4 engineered features: elo rating, recent form, head-to-head, and toss impact. blended with official ICC rankings (70/30 split).

full-stack AI tool that accepts CSV data as input and automatically generates a clean python data-cleaning script tailored to the dataset. built with a react frontend and python backend, deployed on vercel.