Side-by-side eval runs
Compare GPT-4o vs. GPT-4.1 or prompt v1 vs. v2 in one place, with outputs lined up instead of buried in logs.
A lighter home for the open-source products I want people to discover, fork, and actually use.
OpenEvals is the first release: an open-source studio for prompt evals, model comparisons, and regression tracking. The goal is simple: make eval workflows feel as approachable and polished as the products they protect.
Prompt evalsRegression baselinesGitHub ActionsOpenAI-first
Regression tracking
Pin a baseline run, rerun the suite later, and see whether the prompt actually improved or drifted backward.
GitHub-native workflow
Suites live as YAML, can be forked cleanly, and plug into CI so the repo works for both solo builders and teams.
This page is now structured like a real catalog, not a one-off feature block.
Project Tabs
One tab per shipped open-source project, with room to grow.
3 live now. As new repositories launch, they can drop into this strip without another redesign.
SystemsDataMLTab 01
BabyGuard — WiFi Breathing Monitor
Contactless baby breathing monitor that turns a $4 ESP32 and any WiFi router into a real-time infant apnea detection system. No wearables, no cameras, no subscriptions — just physics.
Why It Belongs Here
Full-stack IoT system: embedded C firmware, real-time DSP pipeline, alert infrastructure, and 3D visualization.
ESP32 firmware captures WiFi CSI at 50 Hz with real-time gain compensation.
Python DSP pipeline: Hampel filter → bandpass → PCA subcarrier fusion → FFT/autocorrelation breathing estimation.
Apnea detector with alert system: push notifications, audible alarm, and watchdog for silent failure prevention.
PythonCESP-IDFNumPySciPyFlaskWebSocketsThree.js
View BabyGuard — WiFi Breathing Monitor
More project tabs plug in here automatically.
Preview linked to GitHub
Open projectSignal
64 subcarriers × 50 Hz
Detection
<12s apnea alarm
Tests
32 passing
The packaging is meant to help strangers understand the project fast.
Public repo
OpenEvals is already live on GitHub with release notes, discussions, and a contribution surface.
Self-hostable
The stack is React + FastAPI + PostgreSQL + Redis so people can run it locally or deploy it without guesswork.
Built to be shared
Example suites, GitHub Actions support, and clean screenshots make it easier for people to post, fork, and benchmark publicly.
This page will stay separate from the main portfolio and grow into a focused shelf of products.
OpenEvals is the first project here because it already has a real release surface, a clear use case, and a reason for strangers to share it. More open-source systems and ML projects will land here over time.
Useful on first open
A repo should explain itself quickly. People should understand what it does, why it matters, and how to try it within minutes.
Product polish matters
Good OSS should feel intentional, not like a dumped side project. Presentation is part of adoption.
Designed for community loops
Projects here should invite comparison screenshots, forks, issues, and benchmarks instead of quietly sitting in a portfolio grid.