mentor-worker-benchmark
AI-assisted experiment / Python
Mentor-worker benchmark
An AI-assisted benchmark experiment around whether a mentor LLM can improve a worker model on deterministic coding tasks.
Selected Work
Six experiments that explain the profile.
AI-assisted prototype / TypeScript
TaskProof
An AI-assisted browser proof prototype that turns a task spec into inspectable evidence bundles and static reports.
AI-assisted analysis experiment / TypeScript
Boundary Atlas
An AI-assisted experiment in TS and JS architecture analysis with findings, drift checks, and offline HTML reports.
AI-assisted testing experiment / TypeScript
Counterexample Studio
An AI-assisted experiment in making property-based counterexamples, shrink traces, and reruns easier to inspect.
AI-assisted creative prototype / TypeScript
PixelMelt
An AI-assisted browser prototype that turns one image into a live falling-material simulation and exportable WebM clip.
llm-eval-harness
AI-assisted CLI experiment / Python
LLM eval harness
An AI-assisted CLI experiment for reproducible local-LLM evals over any OpenAI-compatible API.
Other Public Work
Other prototypes worth seeing.
PromptOps Studio
AI-assisted full-stack prototype for text generation, summaries, titles, and keyword extraction.
Open repoMed Explain
AI-assisted local prototype for explaining clinical PDFs with OCR fallback, lab parsing, and FHIR export.
Open repoTrolley Tracker
AI-assisted campus transit prototype for live trolley positions, stop ETAs, rider alerts, and schedule fallback.
Open repoHow To Read This
This is a prototype portfolio, not a claim of mastery.
AI-assisted from end to end
I use AI heavily for coding, iteration, and exploration. What is actually mine here is choosing ideas, steering the shape, testing what matters, and polishing the result.
Prototype-heavy by design
Most repos are experiments or tool concepts pushed far enough to become believable demos. They should not be read as proof that I have deep experience in every domain they touch.
Packaging is the skill
The throughline is fast iteration, sharper demos, cleaner docs, and making weird ideas legible enough that someone can actually try them.
Benchmark
mentor-worker-benchmark
An AI-assisted benchmark experiment around a question I found interesting: whether a mentor LLM can improve a worker model on objective coding tasks. The real signal is the artifact trail, not any claim that I am a benchmark researcher.
Browser Proof
taskproof
An AI-assisted prototype for turning a browser task spec into screenshots, DOM captures, network evidence, console logs, and a shareable static report. It is a good example of how I shape an idea into a clean demo surface.
Static Analysis
boundary-atlas
An AI-assisted architecture-analysis experiment for TypeScript and JavaScript repos. I would not present this as evidence that I am an architecture expert. I would present it as evidence that I can turn an analysis idea into a polished tool-shaped prototype.
Testing UX
counterexample-studio
An AI-assisted experiment focused on the part of property-based testing that usually feels opaque: the minimal failing witness, the shrink path, and the exact rerun command that gets you back there.
Creative Tool
pixelmelt
A more playful AI-assisted prototype: turn one image into a live material simulation, then export a short clip directly from the browser. This is here because it shows experimentation and taste, not because it proves graphics expertise.
CLI Utility
llm-eval-harness
A narrow AI-assisted CLI experiment for sanity-checking a local model server, saving artifacts, and keeping a small eval history while swapping prompts, models, or serving stacks.