What we have built, and what was measured.

The situation. The U.S. Army's xTech|Apex Intercept competition sought fire control and flight guidance that could run inside the cost and power limits of a low-cost interceptor.

The constraint. A guidance loop needs a bounded worst-case latency, not a good average. That rules out the general-purpose compute path before any benchmark is run.

What we did. Implemented the guidance loop in programmable logic on AMD adaptive silicon: seeker fusion, state estimation, proportional navigation and autopilot, with timing as a property of the design.

What was measured. Selected as one of seven Topic 4 finalists and awarded the $25,000 finalist prize. At the finals, the loop ran at 5 kHz on an AMD Kria KV260 with zero measured jitter, above its design target.

Status. Finalist. Winners are named by December 31, 2026.

The situation. Perception and control for small autonomous platforms, where the compute budget is set by the airframe.

The constraint. Everything had to fit one module: perception, sensor fusion and the control interfaces, with no cloud and no companion computer.

What we did. Built the stack on an AMD Kria KV260, with perception in logic and the ROS 2 and MAVLink interfaces on the processor system. Flown on fixed-wing, multirotor, ground and legged platforms, and extended to a five-aircraft swarm.

What was measured. Roughly ten watts for the full stack on the KV260. Preflight checks verify actual data flow rather than parameter configuration, which is the failure that grounds most integrations.

Status. Flight proven and in active use across programs.

The situation. We wanted our own throughput claim to rest on an instrumented program rather than on a vendor benchmark.

The constraint. Throughput claims usually measure generation speed alone. Verification is where the cost sits, and it is the part nobody instruments.

What we did. Delivered a production platform with two engineers on an AI-assisted pipeline, recording engineering hours against delivered source.

What was measured. 41,200 production lines in roughly 575 engineering hours, against a COCOMO II estimate of 170 to 220 person-months. About 72 lines per engineer-hour uncertified, and 15 to 24 under structured verification and validation. We report the two rates separately because the certified band is the one enterprises buy.

Status. Delivered.

The situation. Running a useful language model where there is no cloud, no link and a hard power budget.

The constraint. Weight memory traffic is what makes conventional inference expensive. Removing it means changing the numeric representation as well as the accelerator.

What we did. Built a ternary architecture with a constant recurrent state, so there is no growing attention cache, and wrapped it in a grounded retrieval harness with constrained decoding that cannot answer outside its sources.

What was measured. A 1.3 billion parameter model running on a single AMD Alveo U50, with four deployed knowledge bases covering general, maritime, defense and industrial content.

Status. Deployed and running. Larger models are in development.