William
Lonon.
I'm a data scientist and software developer who thinks in problems and solutions — and likes owning the whole path between them. From raw data architecture to working, ML-powered products, I build useful programs, tools, instruments, and websites. Below are a few of my favorites.
Projects
Data science and software I've built to do real work — machine learning, AI pipelines, and automation that runs.
An open-source methodology and pipeline for auditing bias in multimodal generative AI. It runs GPT-4o and DALL·E in a recursive "telephone game" — chaining image captioning and text-to-image generation across many iterations — and quantifies how bias compounds across cycles using CLIP similarity and facial-recognition metrics to track semantic drift, identity preservation, and information loss.
- Async pipeline chaining GPT-4o captioning and DALL·E generation over many cycles
- CLIP similarity + facial-recognition metrics quantify semantic drift and identity preservation
- Surfaced measurable gaps between stated content policy and real model behavior — up to ~1800% more hedging on a public figure across recursive prompts
End-to-end, LLM-powered outreach pipeline built to promote releases at scale — targeting 9,000+ music-industry contacts with personalized, genre-aware pitches. Born out of running Octoquan Records.
- Playwright scraper handling virtual scrolling and dynamic DOM on Airtable
- Claude API integration for personalized email generation by genre & contact type
- Full pipeline from data extraction to automated delivery
A desktop tool that turns any audio file into a playable sampler instrument. Point it at a recording — a drum loop, a vocal take, an engine — and it slices the audio at every onset, fingerprints each slice by timbre, and maps them onto keys — either by clustering acoustically similar sounds or classifying them into labeled drum types. Pick a file, hit Run, and play it with your keyboard.
- Onset slicing + timbral feature extraction with librosa
- Unsupervised clustering: KMeans, Agglomerative, and HDBSCAN
- Supervised k-NN classification into labeled drum classes (kick, snare, hat…)
- Real-time playback engine with ADSR, filtering, and velocity crossfade
Description In, Resume Out. Feed DIROBot a job description and your background, and it generates a tailored, ready-to-send resume. Currently in development — live demo soon.
Sites I've Built
Live sites I've designed and shipped — for myself and for others. Hand-built, deployed, and running in production.
Independent music label I founded and built from the ground up — a home for artists making unconventional music while keeping ownership of their work. Minimalist, underground-leaning design.
Step-by-step LibGuides I author for the University of Arkansas Libraries' AR Creation Studios — official manuals that let anyone self-serve the space, from first-timers to pros. The flagship is the Audio Recording Studio guide, written end to end: equipment, core audio concepts, the full recording workflow, troubleshooting, and policies.
A landing page for the University of Arkansas Experiential Learning + AR Creation Studios — the Events Center, Makerspace, Recording Studios, Meeting Rooms, and Immersive Media spaces. Designed and built to give the program one clean, shareable home.
About
I'm a developer and data scientist who likes to own the whole stack — from the model and the data to the interface someone actually clicks on.
Most of what I build starts as a problem I personally hit: a label that needed outreach, a site that didn't exist yet, a workflow worth automating. I'd rather ship a real thing than talk about one.
This site is a small example — no framework, no page builder, just hand-written HTML, CSS, and JS deployed on Vercel.
something.
Open to development work, collaborations, and full-time roles. Need a site or a tool built? Reach out.