Stefan Webb

Oumi - Build and deploy custom AI models from a prompt in hours

by
The right model beats the biggest model. Build and deploy custom AI models from a prompt in hours — with 50% higher accuracy, 10× lower cost, and 10× lower latency than general-purpose LLMs.

Add a comment

Replies

Best
Stefan Webb
Maker
📌
Before Oumi, our CEO, Manos Koukoumidis, led the effort behind Google Cloud PaLM — coordinating 300+ contributors across 30 teams to build one of the most advanced AI systems ever made (what became Gemini). And he kept seeing the same anti-pattern: Every enterprise running AI in production was using the same massive, generic model behind a proprietary API for completely different jobs. Custom models were always the answer. But building one took months, a full ML team, and deep expertise most companies don't have. So we built Oumi. Just describe what model you need, starting with just a prompt – Oumi handles everything; evaluation, data synthesis, training, and continuous improvement. What will you build with Oumi? We'd like to hear about your custom AI use cases in the comments bellow.
Konstantinos Anastasiou

In our days that energy costs are so important you don't need a model that will burn tokens meaningless. You need the tokens spent to address each enterprise's needs i.e. A token trained and fit customized for each enterprise. This is what Oumi's product accomplishes in hours - not months.

Min Song

Hi all, I'm super stoked to share with you all our platform that helps anyone to be able to create custom models of your own! Based on the vast choices of open source models you can find, users can train synthesize and evaluate your own model based on your own use case :) Just try it once and do a full loop of, synthesizing a dataset, training a model, evaluating how good your model is for your use case, then synthesizing again, and seeing your accuracy improve.

It's super exciting to share a platform that has a huge potential of being the default way of building models, removing so much friction for anyone to have their own custom small models.

Let us know what crazy model you want to build, and I challenge you to also try breaking our service!

Timothy Leffel

It's been a wild ride working on the Agent interface and training recipes for the Oumi platform! Every base model has its own quirks, and every task is different, so a lot of work goes into enabling each entry in the Base Model dropdown. We'll continue onboarding new open-weight models to training as they're released :)

Would love to hear which base models people are most interested in using!