Speaker
Description
"High-energy physics experiments are increasingly using FPGAs to deploy machine learning models for low-latency triggering and real-time data processing. Deploying these models on FPGAs requires coordinating several software and hardware tools. We present an agent workflow using reusable skills to support this process. The agent coordinates model conversion, hardware builds, and on-board testing, while scripts execute the tasks and check the results.
We have used the workflow to deploy multiple models on two FPGA boards. Initial validation includes comparisons with software reference outputs and checks of resource use and timing. Further work will test consistency across boards and independent reproduction, and compare the workflow with a generic agent to evaluate deployment success, execution time, and the need for human intervention."
