Oumi: Fast Custom Models
Oumi launched a no‑code product promising custom AI models 'in hours,' with the CEO stressing iterative retraining on failure patterns to reach production quality — pitched as a rapid way to build deployable models without heavy engineering. The tool appears aimed at teams needing quick model prototyping and retraining workflows. (x.com)
Emmanouil “Manos” Koukoumidis is Oumi’s CEO and co‑founder, and he previously led Cloud PaLM efforts at Google Cloud and worked on early NLP and RAG prototypes during a multi‑year tenure in industry research. (oumi.ai) Oumi closed a $10 million seed round announced January 29, 2025, with Venrock and Obvious Ventures named as lead investors and participation from Plug and Play and Ascend Venture Capital. (ascend.vc) The company’s public website advertises specific performance claims — “50% higher accuracy, 10× lower cost, and 10× lower latency” on task‑specific models — and offers $50 in trial credits while positioning customers to run models on their own infrastructure. (oumi.ai) Oumi’s codebase and tooling are publicly maintained on GitHub under an open source license, with thousands of commits, recent upgrades for Transformers v5 and vLLM compatibility, and an active repo for fine‑tuning, evaluation, and deployment workflows. (github.com) Product documentation and previews describe an automated lifecycle that generates test sets and evaluators from a task description, recommends metrics, and provides a deploy command and integrations (including recent deploy support for fireworks.ai and parasail). (aitoolsclub.com) Oumi is organized as a Public Benefit Corporation based in the Seattle area and was founded by engineers with backgrounds at Google, Microsoft and Apple, including co‑founders Panos Achlioptas and Konstantínos Aísopos alongside Koukoumidis. (oumi.ai)
