Skip to main content
Prompt learning uses evolutionary algorithms to automatically optimize prompts for classification, reasoning, and instruction-following tasks. Use the checklist below as your map; follow the linked pages for implementation detail.

​
1. Build a prompt evaluation task app

Define a task app that evaluates prompt performance on your task (classification accuracy, QA correctness, etc.).
→ Read: Task App requirements

​
2. Deploy and verify the service

Smoke-test locally, then deploy to Modal or your host of choice once health checks pass.
→ Read: Deploying task apps

​
3. Author the prompt learning config

Capture algorithm choice (GEPA or MIPROv2), initial prompt template, training/validation seeds, and optimization parameters in TOML.
→ Read: Prompt learning configs

​
4. Launch the optimization job

Run uvx synth-ai train --config config.toml to create the job and stream status/metrics.
→ Read: Launch training jobs

​
5. Query and evaluate results

Use the Python API or REST endpoints to retrieve optimized prompts and evaluate them on held-out validation sets.
→ Read: Querying results