Travis Muhlestein PRO
TravisMuhlestein
AI & ML interests
Product & AI CTO at GoDaddy focused on AI infrastructure, orchestration, agent systems, observability, and enterprise-scale AI deployment
Recent Activity
posted an update 4 days ago
Building a multi-agent system for automated data pipeline generation
We recently built a conversational agent to automate a recurring data engineering workflow: creating and modifying data pipelines.
The architecture uses a manager agent to classify user intent and route requests to two specialized agents:
Pipeline Creation Agent: inspects S3 schemas, generates PySpark jobs, creates Airflow DAGs and environment-specific policy files, and opens pull requests.
Column Addition Agent: locates existing pipelines, identifies downstream dependencies, modifies pipeline scripts, and creates pull requests.
The system is exposed through a FastAPI service, with Slack, REST API, Swagger UI, and GitHub Actions as entry points.
One architectural decision made a significant difference: knowledge-base-first generation.
Initially, we generated code and validated it against our Lake Naming Standards afterward. That resulted in corrective commits and unnecessary PR updates.
We changed the sequence: each specialist retrieves the relevant standards before generating code. This puts naming conventions, database structures, and file paths into the model's context upfront, rather than treating compliance as a post-generation check.
The workflow went from 2–4 hours of manual work to a few minutes.
A few areas worth exploring further:
How should specialist agents divide responsibilities?
How much deterministic validation should sit outside the LLM?
How do we make generated code compliant by design, rather than relying on correction loops?
Full implementation details:
https://www.godaddy.com/resources/news/building-an-ai-agent-for-data-pipeline-creation posted an update 24 days ago
I've been interested in how coding agents handle tasks beyond generating code.
Deployment is a useful example: it involves interacting with a real environment, following a sequence of steps, and checking that the result actually works.
The GoDaddy Node.js hosting workflow supports deployment through its CLI, including workflows that can be driven by a coding agent.
The walkthrough covers the setup and deployment process:
https://www.godaddy.com/resources/news/deploy-a-node-js-app-to-godaddy-from-your-terminal-or-let-your-coding-agent-do-it
Curious how others are handling deployment in agent-driven workflows. Which steps do you let the agent execute, and which do you keep under direct human control?
#CodingAgents #NodeJS posted an update about 2 months ago
As AI agents become more autonomous, identity becomes part of the infrastructure problem.
We’ve been working on Agent Name Service (ANS) to give agents a verifiable identity before they interact or exchange sensitive information.
One design goal is that verification shouldn’t depend on a centralized service for every request. ANS can verify an agent offline against published root keys, with the verification path completing in under 0.5 ms on a single server core.
The work builds on infrastructure we already have around domains and DNS.
We’re also looking at that infrastructure at the DNS layer through SVCB and HTTPS records, evolving how modern applications connect beyond traditional CNAME-based architectures.
Two engineering deep dives:
→ Don’t Trust. Verify. Offline, Sub-Millisecond Agent Verification with ANS
https://www.godaddy.com/resources/news/dont-trust-verify-offline-sub-millisecond-agent-verification-with-ans
→ Beyond CNAME Flattening
https://www.godaddy.com/resources/news/beyond-cname-flattening Organizations
