<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Grounded AI</title><description>Josh Longenecker on evaluating AI systems, decision models, and other things worth measuring.</description><link>https://groundedai.tech/</link><item><title>Hardware entitlement to roofline: a 7-day curriculum</title><link>https://groundedai.tech/blog/hardware-entitlement-roofline-curriculum/</link><guid isPermaLink="true">https://groundedai.tech/blog/hardware-entitlement-roofline-curriculum/</guid><description>Take an op and a chip, work out from the documentation how fast the op can run, measure how fast it does run, and explain the difference. Seven days, built on AWS Trainium and Inferentia2, with a check-yourself problem for each.</description><pubDate>Sun, 11 Oct 2026 00:00:00 GMT</pubDate><category>trainium</category><category>nki</category><category>roofline</category><category>hardware</category></item><item><title>Tensor Engine clockwork: one matmul on Trainium, cycle by cycle</title><link>https://groundedai.tech/blog/tensor-engine-clockwork/</link><guid isPermaLink="true">https://groundedai.tech/blog/tensor-engine-clockwork/</guid><description>An interactive walk through what the Trainium/Inferentia Tensor Engine does on every clock of one nc_matmul: weights parked in a systolic array, inputs streamed in, partial sums falling into PSUM. Plus how the same picture shows up in a Neuron Explorer profile.</description><pubDate>Sat, 10 Oct 2026 00:00:00 GMT</pubDate><category>trainium</category><category>nki</category><category>roofline</category><category>hardware</category></item><item><title>Half the latency, 61% cheaper: cascading a decision model in front of an LLM</title><link>https://groundedai.tech/blog/cascade-evaluator-tool-calls/</link><guid isPermaLink="true">https://groundedai.tech/blog/cascade-evaluator-tool-calls/</guid><description>CascadeEvaluator lets TypeSafe’s Jev settle the decisions it is sure of and passes the rest to any LLM judge you pick. Against Claude Sonnet 4.6 on 200 real agent turns, it was 51% faster and 61% cheaper, with no loss in accuracy.</description><pubDate>Thu, 08 Oct 2026 00:00:00 GMT</pubDate><category>evaluation</category><category>decision models</category><category>agents</category><category>grounded-ai</category></item></channel></rss>