A working subgroup of the ISoP AI/ML SIG, bringing pharmacometricians together around AI agents and agentic workflows.
Two ways in, both take under a minute. Membership is free and open to anyone.
- Sign up with the form. Quickest way in.
- Subscribe by email. Send a blank message; works behind corporate firewalls.
Once you are in, the conversation happens in Discussions.
New here? Try PMxbench: one simulated population PK study with a known answer. Analyze it by hand or with an AI agent and send back one YAML file. You don't need a good score. A submission that says what blocked you counts.
To bring together a community of pharmacometricians to discover, evaluate, and produce agentic workflows, codifying their collective expertise into shared, verifiable standards of practice that grow with the field.
Three objectives, three lanes:
- Community & shared learning. Bring together pharmacometricians interested in agentic AI: unify terminology, lower the barrier to entry, and learn in the open.
- Discover & harmonize tools. Curate the community's best agentic tools into one open, readable stack the group can reuse and grow.
- Evaluation & trust. Benchmarks and tasks that evaluate PMX agents, plus the QC, traceability, and human-in-the-loop practices that turn scores into trust and make workflows regulatory-ready.
How they connect: learn by doing, tools discovered and evaluated, evaluation turned into trust.
The group publishes at aiml-sig.github.io/Agentic-workflows. Three things worth going straight to:
- PMxbench. The landing page: get the data, run the analysis, submit. The files live in
tools/pmxbench/. - Glossary. Plain-language definitions of the terms this field uses loosely: agent, harness, tools, MCP. Start here if the vocabulary is the barrier.
- Leaderboard. PMxbench scores for every recorded submission: which workflow, on which model and harness, at what score and cost. To add a row of your own, submit a result.
Released under the MIT License.