avodaq Dynatrace Agent Assistant

Inspiration

We wanted to eliminate that friction by letting user simply describe what they want, and have an AI agent handle the rest, using their own existing Monaco repo as the source of truth.

What it does

An AI-powered configuration assitant that reads your current Monaco config as code, ask the questions, construct the correct settings payload, all based on written best practices.

How we built it

With Google ADK as the multi-agent orchestration framework, Gemini 3 Flash as the LLM backbone. Three stage agentic pipeline: planner agent -> planner_review_agent -> settings_validation_agent.

Challenges we ran into

SKILL.md writing skill in a way that agent could understand efficiently. Modeling the stateful HITL loop within Google ADK's workflow required careful edge design. Optimization on agent instruction to ensure consistency and avoid hallucination.

Accomplishments that we're proud of

A fully working end-to-end pipeline that takes a natural-language requests and produces a vlidated Dynatrace Settings 2.0 payload.

What we learned

Structure skills are far more reliable than cramming instructions into system prompts. HITL patterns in agentic workflows that is better than assumption of best answer.