Master any domain with a custom AI coach

Kampé generates a weighted rubric that coaches your agent through structured evaluation — then export it to Claude Code, ChatGPT, OpenClaw, or train right here.

What should your AI be great at?

Try an ignite

Example coach output

Kidney Transplant Caregiver Coach

Medical Accuracy30%

Correctly identifies conditions, contraindications, and evidence-based treatment protocols.

Emotional Support25%

Acknowledges patient/caregiver anxiety and provides reassurance grounded in facts.

Coach: Tendency to jump to clinical details — lead with empathy first.

Insurance Navigation20%

Guides through pre-authorization, appeals, and coverage verification processes.

Care Coordination15%

Connects care steps across specialists, labs, and scheduling dependencies.

Risk Communication10%

Explains procedural risks in plain language without minimizing or alarming.

Forged by Kampé

Forge your coach here. Deploy it everywhere.

OpenAI

Anthropic

Grok

OpenClaw

JSON

Export as CLAUDE.md, Custom GPT instructions, SKILL.md, or raw JSON — your coach works wherever your agent lives.

Describe

Tell us what your AI needs to be great at. Smart intake asks 2-3 questions to sharpen the rubric.

Coach forged

We generate a domain-specific rubric with weighted criteria and practice scenarios.

Export anywhere

Download your coach for Claude Code, ChatGPT, OpenClaw, or train right here.

78→88.5

In testing, agents coached with Kampé rubrics improved from 78 to 88.5 on domain-specific tasks. Not because they got smarter — because they got coached.

Research confirms this — “Rubric Is All You Need” (ACM 2025) and rubric-based RL rewards (ICLR 2025) show structured evaluation produces measurably better results than freeform feedback.

How is this different from just prompting Claude?

Prompting an LLM cold

One-shot response, no evaluation structure
No memory of what worked or failed
Generic to the model's training data
No way to measure improvement

Kampé-coached agent

Weighted rubric with domain-specific criteria
Iterative feedback loop with trajectory tracking
Coaching notes that accumulate across attempts
Human domain knowledge injected via Power Boost
Measurable score improvement over time