Çağatay Çallı
Senior ML/AI Engineer · LLMs · Search · Recommendation Systems
A few things about me
- Started in R&D and text mining - semantic search, NLP, a thesis on Explicit Semantic Analysis and search result clustering, and Google Summer of Code at the Eclipse Foundation.
- Co-founded a social media analytics startup.
- Helped scale Udemy, focusing on search, recommendations and ML infrastructure.
- Built cloud data platforms and Spark pipelines at bp across globally distributed teams.
- Shipped the DuckDuckGo Android app early on.
- Now independent, I help teams design and ship LLM, search and recommender projects end-to-end.
Selected Projects & Publications
SpeechLens — Fast on-device speech enhancement on Mac running NVIDIA's RE-USE model.
macOS menu bar app built with Swift, MLX and custom Metal kernels. On Apple M1 Max, SpeechLens enhances stereo audio faster than real time at 16, 44.1, and 48 kHz.
Benchmarks · Download for macOS
"EasyESA: A Low-effort Infrastructure for Explicit Semantic Analysis" - ISWC 2014
Wikiprep-ESA - Open-source Explicit Semantic Analysis (Python, Lucene)
Ekşin - Android client for Ekşi Sözlük, Turkey's largest social dictionary (30K+ users)
Google Summer of Code 2007 - The Eclipse Foundation (Eclipse Platform & ECF)
What I work on
Philosophy
- Most ML problems are product and data problems in disguise.
- Simple, well-instrumented models in production beat fancy models stuck in notebooks.
- Owning the full loop (data → model → serving → eval) matters more than owning a narrow slice.
- Clear metrics and honest evaluation save more time than any clever architecture.
Education
Thesis: "Improving search result clustering by integrating semantic information from Wikipedia"