A metric that knows good from bad
GEPA learns from feedback on every output. Someone has to encode your quality bar into a function that scores and critiques before any optimization starts.
GEPA, fully managed.
Bring an unlabeled dataset and a one-sentence task. Sutro surfaces the hardest cases for your experts, learns your decision rules from their feedback, and hands you an optimized, deployable AI Function.
GEPA works. Running it is less straightforward.
GEPA is the state of the art in automatic prompt optimization. Instead of hand-tuning, it evolves prompts: trying variations, reflecting on failures in plain language, and keeping what works. The results are real: in the original paper it beats DSPy’s default optimizer and even RL fine-tuning, at a fraction of the compute.
Using it in production is another story. Before GEPA can optimize anything, you have to build everything around it:
Labeled examples it can trust
The search is only as good as its examples. Which cases are worth labeling? Are the labels right? One mislabeled edge case quietly steers every prompt that follows.
A harness to babysit
Reflection models, rollout budgets, retries, rate limits. It’s configuration, not judgment. And it’s where the weeks go.
Building an AI Function isn’t about scaling general intelligence. It’s about teaching a model exactly how your organization wants a task performed. Sutro turns GEPA-style optimization into exactly that: you teach the task on the hardest examples, and we handle the rest.
Focus on scaling your judgement, not tweaking tooling.
Run it yourself
- Find and annotate the right mix of high-value cases
- Design the metrics and judges that score every prompt candidate
- You choose the reflection model and tune the budgets
- You own a pipeline, and its maintenance
- Great for research and custom multi-module programs
Let Sutro run it
- Bring unlabeled data and a task description
- Annotate only the cases where models disagree
- Optimization and infrastructure managed end to end
- You get a deployable, portable AI Function
- Great for classifiers, extractors, and judges your team must trust
“Sutro saves our team countless hours, and it gives us the invaluable ability to measure, optimize, and prevent regressions against our domain expertise. It’s a must-have for serious AI developers.”