Things I've built. A mix of personal projects and course work. You can filter by tag or conduct a search.
Personal
A free, open-source AI resource hub for everyday professionals. Built for the person who hears about AI constantly but doesn't know where to start — and doesn't want to wade through hype to get there.
Includes a plain-English crash course, five deep-dive articles on LLMs, RAG, agents, and more, a searchable library of 15 ready-to-use prompt templates, six milestone projects organized by skill level, a curated directory of 26 AI tools across 8 categories, a glossary of 30 AI terms, and an FAQ. Single HTML file, no framework, deployed on GitHub Pages.
Paste a job description, get a tailored resume back. It rewrites your bullets to match the JD's language, updates your summary and skills section, flags ATS issues, scores keyword overlap and honest fit separately, generates a cover letter, an elevator pitch for phone screens, and maps 3–5 STAR stories to likely behavioral questions for that role.
Built on Claude's API. Everything runs in one call. Results are saved to a history tab — every run stored with company, title, date, and scores so nothing gets lost between sessions.
A full guitar learning environment organized around four phases — Absolute Beginner through Advanced. Each phase has a structured curriculum, milestone songs with tutorials, gear recommendations, and exit criteria so you actually know when you've leveled up.
Beyond the curriculum: a skill tracker (44 skills, each with curated resources), a practice timer that builds a session plan from your available time, a metronome built on the Web Audio API with tap tempo and lookahead scheduling, a chord library (37 SVG diagrams), a scale library (9 scales across all 12 keys with full neck diagrams), a practice log with BPM tracking and streak counter, and a placement quiz that drops you into the right phase based on your current level.
UT Austin — Generative AI for Business Applications
Projects built as part of my GenAI graduate coursework.
Manual NASDAQ monitoring is slow and biased — two analysts reading the same headline can reach opposite conclusions. We built an automated sentiment classifier to replace that with objective, real-time signals.
We evaluated four model architectures and the Sentence Transformer came out on top: 77.8% accuracy, +9% F1 improvement over Word2Vec. Input is a financial news headline or article; output is a sentiment score — positive, neutral, or negative.
The Merck Manual is 4,104 pages. No clinician can hold all of it — especially under pressure. We built a RAG system on top of GPT-4o that ingests the full manual into a vector database and returns grounded, source-cited answers to clinical questions.
Every response includes a page citation, so the output is verifiable rather than generated from thin air. We benchmarked three approaches — raw GPT-4o, prompt-engineered GPT-4o, and the full RAG pipeline — and scored outputs using an LLM-as-a-Judge framework across groundedness and relevance.
Developed a functional GenAI chatbot for a food aggregator to automate high-volume customer inquiries. The system utilizes an LLM-driven SQL agent to query structured databases in real-time, providing immediate updates on order status, payments, and delivery tracking.
Used prompt engineering and input/output guardrails to automate real-time database querying and natural language response synthesis.
No projects match the current filters.