EH-rik ยท keen-tuh-NEE-yah
LinkedIn / Twitter / GitHub / Google Scholar
Co-founder of refresh.dev (YC X25) working on RL gyms for coding & computer use.
Based in San Francisco, CA.
Formerly lead MLE at LynkAI (large scale voice data, models).
Previously early engineer at Superdial working on IVR navigation.
Previously large scale data at Capital One, scraping at Amazon; submitted a 2021 Amazon patent on automatically generating scrapers.
Previously vision researcher working on GAN-based user preference modeling and cross-view synthesis (ACM-MM '20, ECCV '20).
In college, worked on the Citadel Aperture program doing financial modeling; then on the Numerai prediction modeling competition (2x Gold, 99th percentile; 5x Silver in Numerai Fund's Weekly AI/ML Competition). Worked on a synthetic data generation project that improved model performance with GANs.
Won hackathons hosted at UChicago, Vanderbilt, and Amazon.
You Can’t Round-Trip Audio Through Log-Mel: why audio to mel to audio is lossy, why vocoders exist, and when the gap matters.
Recoverable Before Readable: an endpoint-coordinate lens on transformers, reading the residual stream as a trajectory to catch jailbreaks early, audit interpretability probes, and watch decisions form (Qwen3.5-2B).