Jordan Davis

Three things rarely coexist in the same engineer: defense-grade audit-bound delivery, first-hand agentic AI systems design, and native real-time ML. Jordan sits at that triangle.

By day he is the technical lead on an Exostar SaaS product serving tier-1 defense primes — compliance-bound, review-driven, traceable. By night he runs Tensorpunk, the lab that built Relay (the #1 ranked system on a public AI-memory retrieval benchmark) and Retro (a programming language designed for LLM-driven development). The two streams reinforce each other: agentic primitives invented at Tensorpunk Labs are translated — with prompt-injection guards, query safelisting, and audit logging — into the defense-grade product at Exostar. That translation is the whole differentiator.

He ships end-to-end: model training → ONNX export → C++ runtime → UI → deployment → audit trail. No handoff seams.

Jordan is also a working electronic musician who records and performs as TENSOR. Tensorpunk was founded at the intersection of that practice and his engineering work — it is the name of both the AI audio company and the artist project that drives its R&D agenda. MACE, QuiverAPI, and ANVIL exist because Jordan needed them for his own sound design and production; they graduated into commercial products once professional artists — including Grammy winners — adopted MACE for real sessions.

This creative lineage is not decorative — it shapes how he engineers. Audio plugins have to run at sub-10ms latency inside a DAW on a musician's laptop. The C++ runtime, ONNX deployment, and CUDA / CoreML acceleration work transferred directly into the kind of real-time, resource-constrained, reliability-sensitive systems mission-critical software demands. End-to-end model ownership (train → export → deploy → UI) is a habit, not a project plan.

In 2025 he collaborated with Drone Machines (Tristan Shone / Author & Punisher — a veteran underground electronic artist known for building custom industrial MIDI hardware) to design a cross-platform JS / Electron config UI that lets non-technical musicians customize his performance gear. The undergraduate degree is from the University of Cincinnati College-Conservatory of Music — an institution more often associated with classical performance than systems engineering. That dual-literacy shows up in how Jordan designs interfaces and documentation: musician-legible, engineer-correct.

The creative practice is why Tensorpunk Labs exists in the shape it does. The agentic developer tools (Relay, Retro, Lattice) are infrastructure Jordan built to coordinate the R&D lab running behind the audio products — then spun out as open source once they proved more broadly useful.

Continuous defense / aerospace exposure since 2022, across three roles, with the tempo and review culture of regulated environments internalized.

Exostar (2024–present) — technical lead on a tier-1-prime-facing SaaS platform. Compliance-bound, audit-driven. Robot Morning (2023–2024, acquired by Exostar Dec 2024) — Senior Product UI Developer; team and product rolled into the new Exostar platform at acquisition. GE Aerospace (Oct 2022 – Mar 2023, contract) — C++ / Python visual inspection for aerospace part defect detection; reverse-engineered undocumented codebases, implemented image-analysis algorithms against real production parts, designed 3D UI / UX.

Three design commitments that map directly onto regulated environments:

1. Traceable, review-friendly AI. Every AI-assisted change is auditable. Prompt-injection guards, query safelisting, and audit logging are first-class, not retrofitted.

2. Context over cleverness. Production gains come from context engineering — custom MCP servers, skills, RAG over the real codebase — not from swapping models. The 5x productivity lift at Exostar is a context-engineering result, not a model-capability result.

3. Full-stack integration. Models do not ship alone. Jordan owns training → ONNX export → runtime → UI → deployment → compliance review. No handoff seams for drift to hide in.