Independent AI and machine learning consultant in Cedar Park, TX (Austin metro). Ph.D. in physics, University of Notre Dame. 12+ years building production AI and machine learning, eight of them at Oracle, the last three in its AI Center of Excellence.
jmh-datasciences.com · LinkedIn · joe.hahn@jmh-datasciences.com
- Fixed-Price AI Build: one concrete AI job, built on your own data and tested on your real examples before you rely on it.
- Senior AI Help: hands-on help, by the hour, for teams already doing AI with more work than hands.
- geo-herd-rider: eight LLMs graded blind on ~4,500 agent decisions; the most expensive finished last (write-up).
- diplomacy-A2A: seven Claude agents negotiate, ally and betray through a custom agent-to-agent protocol; a cost-vs-capability benchmark.
- chicago_crime_forecast: 1,000 forecasts a month ahead, R² 0.96 at one month, built with Claude Code.
- prospect-scanner: an LLM judge scored against a simple formula (write-up).
- demand-on-demand: plain-English demand forecasts from a single Claude agent and a fixed ML harness.
- Oracle: in-database RAG and NL2SQL; fine-tune and deploy an open-source LLM on GPU.
The N-body and planetary-ring code further down this account comes from my astrophysics research: 30+ peer-reviewed papers.