high school sophomore interested in AI and finance. i like building tools that solve real problems for real people.
most of my projects come from noticing someone struggling with something and thinking "i could probably automate that." sometimes i'm right, sometimes i learn the hard way why it's harder than it looks.
AI + finance tools:
- invoice-payment-matcher: matches bank deposits to invoices using subset-sum algorithms. built this after watching a small business owner spend 4 hours matching payments manually. 102 tests.
- smart-expense-categorizer: categorizes bank transactions into IRS Schedule C tax categories. uses keyword matching + LLM. helps freelancers stop losing money on missed deductions.
- currency-classifier-cnn: CNN that identifies banknotes from photos, built as an accessibility tool for visually impaired users. trained on real data, 96.7% accuracy on test images. has voice output.
AI tools:
- ai-lease-parser: extracts structured data from lease PDFs using OCR + regex + LLM. outputs clean spreadsheets instead of making landlords retype everything. 173 tests.
- ai-scam-site-detector: analyzes websites for signs of AI-generated content and scam patterns. checks text heuristics, domain info, page structure. my most unique project.
i use Claude as a development partner and i'm transparent about it in every repo. i think being honest about AI assistance is more important than pretending i did everything from scratch. claude helps me move faster and learn deeper. i make the design decisions, pick the problems, and own the code.
all my projects have real test suites (580+ tests total across repos), not just demos.
python, pytorch, pandas, streamlit, scikit-learn. learning more every day.