AI · models · datasets · research
Most of my AI work comes back to one question: how do we make this technology work for people who don't fit the default? For me that means English speakers in Israel who mix in Hebrew, people who'd rather speak than type, and learners who want AI to teach them rather than just answer.
Speech and text models that cope with "Hebrish" — the English-with-Hebrew-words that English speakers in Israel actually use. Off-the-shelf ASR mangles words like teudat zehut or mashkanta; I build the datasets and fine-tunes that fix that.
9 related projects ↓Whisper fine-tunes, ASR evaluations and small models that run locally on a phone. Most of my writing starts as voice, so I care about accuracy on real voice notes, background noise and fast on-device inference.
15 related projects ↓Using AI to make technology easier to use: voice-first input instead of typing, TTS that pronounces Hebrew words correctly, and tools that reduce friction for people who work in more than one language.
8 related projects ↓AI that teaches rather than just does: tutor-style agents, learning loops for code generation, and My Weird Prompts, a podcast that explains a new topic every day.
2 related projects ↓The work I'd point you to first. Items with a card page on this site open it; the rest go to Hugging Face or GitHub.
Parakeet TDT 0.6B v2 fine-tuned for Hebrew words in English speech. Runs offline on Android and Linux; halves code-switching WER with no English regression.
SpaceA replicable benchmark and leaderboard: stock and fine-tuned speech-to-text scored on how many Hebrew words in English speech they get right. My Parakeet fine-tune: 66%; Whisper Large V3: 46%; stock Parakeet: 33%.
ModelWhisper Large v3 Turbo fine-tuned for English speech with Hebrew words mixed in, the way English speakers in Israel talk.
SpaceTry the Hebrish model in the browser and compare it with stock Whisper.
DatasetEnglish sentences with Hebrew words written in Latin letters, labelled at token level (BIO). It trains a classifier that tells TTS which words to pronounce as Hebrew. No comparable corpus existed when it was built.
GitHubCorpus preparation and training code for the Hebrew-in-Latin token classifier.
DatasetSingle-speaker read-aloud audio of English sentences containing Hebrew words, for ASR fine-tuning.
DatasetThe training set behind Whisper Hebrish: English sentences with everyday Hebrew vocabulary.
DatasetRead-aloud English and Hebrew sentences for fine-tuning Parakeet and other ASR models.
DatasetDeveloper and AI vocabulary read aloud, for ASR that has to get technical terms right.
CollectionPersonal Whisper fine-tunes from tiny to Large v3 Turbo, trained on my own voice.
CollectionWhisper fine-tunes with ACFT, sized for on-device use in FUTO Keyboard and other Android apps.
CollectionASR models I have tested running locally on an Android phone (OnePlus Nord 3, CPH2493).
GitHubAn evaluation matrix of open-source Android voice keyboards and voice IMEs.
DatasetReal voice notes recorded in rough conditions, for testing ASR beyond studio audio.
DatasetAnnotated recordings that measure how speaking rate and background noise affect ASR accuracy.
DatasetPairs for fine-tuning a model to clean raw speech-to-text output into readable text.
GitHubAn agent configuration that guides a learner through a problem instead of doing it for them.
GitHubA multi-agent code-generation setup with an educational loop built in.
Every model on my Hugging Face profile, most recently updated first. Most are Whisper fine-tunes for speech-to-text. The main ones have card pages with results, downloads and roadmap.
Every dataset on my Hugging Face profile, split into speech & language work and everything else. The main ones have card pages.
Curated groups of my own and other people's models, mostly around Hebrew and speech.
Podcast
A daily AI-generated podcast. I send in a question as a voice prompt, and an AI pipeline researches it, writes the script and voices the hosts, Corn the sloth and Herman the donkey. Topics run from AI and technology to geopolitics, health and history.
5,862 episodes published as of 14 October 2026, with new ones every day.
Verified by Podtrac (IAB-certified) as of 13 October 2026. Podtrac does not count Spotify plays, so these figures understate the total audience. Show data portal
