Skip to main content

​
Core Capabilities

Sutro turns repeated expert decisions into reliable AI Functions. Upload representative data, review the cases where models disagree, and optimize a prompt against your feedback. Representative data produces model predictions, expert review improves the Function, and the Function runs in Batch or as an event stream.

​
Learn More

Build a Function

Create a judge, classifier, or extractor from your data and decisions.

Run a Function at scale

Execute a published Function over a production list, table, or file.

​
What you can build

  • Evals and judges for agent traces, model responses, conversations, and other quality gates
  • Classifiers and routers for leads, tickets, documents, policy checks, ownership, prioritization, and escalation
  • Extractors and enrichment Functions that turn text, images, and PDFs into typed fields, labels, and normalized attributes
  • Matchers and resolvers for companies, people, products, incidents, and other records
  • LLM-enhanced decisioning that augments classical ML systems for fraud detection, KYC review, spam and abuse prevention, risk scoring, and eligibility decisions with detailed instruction-following and unstructured context
These can be binary decisions, one-of-many or multi-label classifications, rubric-based judgments, or structured extractions. A Function works best when an expert can consistently review its output and when it owns one narrow step in a larger workflow.

​
Run your Function

When the Function is ready, you can efficiently run it offline and at scale using Sutro Batch. You can also copy its optimized prompt and model recommendation along with its output schema into an OpenAI-compatible provider. Sutro is best for repeatable decisions with outputs that can be labeled or verified. It is not intended for latency-sensitive chat or unconstrained long-form generation.