01
Intent & agent systems
PIPER
Connecting intent to task contracts, evidence requirements, and workflow guards. The current Hermes advisory plugin explores one part of a broader architecture.
Explore PIPERIndependent AI Research & Engineering
TechSoft
Systems.
From infrastructure to intelligence.
Building, testing, and understanding AI systems.
From the GPU stack to the agents that run on it.
Tucson, ArizonaIndependent by design.
Inside the lab
Real systems.
Questions worth testing.
Independent applied research in inference, agent behavior, and evidence. Built hands-on, with the methods and limitations kept in view.
All research02
Compute & inference
Beyond model serving
NVFP4 model conversion, vLLM deployment, and workload-specific tuning. Performance research grounded in reproducible tests and explicit limits.
Inside inference engineering03
Research & evidence
Can the result be trusted?
Self-hosted AI-Q research workflows, source traceability, and replayable evaluation. Investigating what changes as evidence moves between agents.
Read the research
Defiantly Not Regex An idea from the lab
Meaning is not
the same as structure.
A visual thesis born from working with models and harnesses. What happens when valid syntax is mistaken for understanding, and what should an intent layer actually do?
Research meets practice
Built on an infrastructure foundation.
TechSoft Systems is led by Johnathan W. Carroll. More than 25 years across enterprise IT, government, ISP operations, and entrepreneurship inform the questions explored in the lab.
Experience & backgroundResearch. Engineering. Real-world questions.