DIRECTOR OF AI ENGINEERING
Remote (Occasional Travel) · U.S. Person Required · Start Immediate
Reports to: Director of Technology
Level: Director
Clearance: U.S. Person required per ITAR 22 CFR 120.15 · Secret-eligible preferred
WE DIDN'T BOLT AI ONTO OUR BUSINESS. WE BUILT THE BUSINESS AROUND IT. WE NEED THE ENGINEER WHO OWNS WHETHER IT'S ANY GOOD.
We are a defense electronics holding company operating across:
• Avionics maintenance and repair • Government contracting across DoD, USAF, USSF, and MDA • Active SBIR/STTR R&D in hypersonic thermal protection, inertial sensors, and advanced optics • An M&A growth strategy across the defense electronics sector
What makes us different:
We run a production fleet of 10 autonomous AI agents as part of our org structure. Not chatbots. Not Copilot add-ons. Each agent has a name, an email address, defined operating authority, a project space, and formal delegation under our governance framework. They execute business development, financial operations, engineering qualification, SBIR program management, and executive coordination — daily, in production.
We are building a fully sovereign, composable, AI-first enterprise platform. IT is invisible. Operations are autonomous. The stack is owned end-to-end — our servers, our encryption keys, our models. No hyperscaler dependencies. No vendor lock-in.
A CAD-integrated LLM running on our on-premises GPU cluster. A government intelligence platform scoring every opportunity that crosses our desk. A formal verification layer that proves — mathematically — that no agent can exceed its delegated authority.
This is in production today.
THE ROLE
You are the model quality owner for the company.
Your mandate: every AI system we deploy — from the government intelligence platform parsing SAM.gov at scale, to the on-premises language model assisting our avionics engineers inside FreeCAD, to the retrieval layer feeding past-work context into every agent's reasoning loop — produces outputs that are accurate, calibrated, auditable, and compliant with ITAR, CMMC, and AS9100.
You are not a researcher. You are not a prompt engineer. You are the engineer who closes the gap between "the model runs" and "the model is right" — and who has the technical depth to actually close it.
You will hire and lead an ML Infrastructure Engineer. You will own a Line of Effort within our operational program. You will report directly to the Director of Technology and interface with the Executive Director on model governance decisions.
WHAT YOU'LL OWN
Government Contracting Intelligence Platform A live .NET/F# application, currently in Sprint 4 of production hardening. Ingests SAM.gov at scale and scores every opportunity against a multidimensional scoring model backed by evidence from an internal truth protocol with provenance hashing and Merkle verification. Score explanations are formally verified via proof-carrying explanations in Isabelle/Coq. You own scoring calibration, hallucination detection thresholds, fact-integrity, and the platform's sovereign migration to our on-premises K3s infrastructure.
On-Premises CAD LLM A 3-node GPU cluster: 6x NVIDIA RTX A4000, 96GB VRAM, running Qwen2.5-7B-Instruct via HuggingFace TGI. It powers an AI assistant built directly into FreeCAD — engineers query it mid-design, in context, with zero data leaving our network. Every thumbs-down feeds a correction corpus. You own the QLoRA fine-tuning pipeline, the feedback loop, and whether this model gets smarter over time.
Retrieval Layer A sovereign IVF-OPQ semantic retrieval system running on K3s that injects semantically similar past work items as context before every agent reasoning loop. A companion IVF-PQ compression layer holds a formally proven Context Retention Rate ≥ 0.844. You own the quality of what agents know before they think.
Model Governance You build the evaluation framework, model change control process, training data governance, and compliance posture — ITAR boundaries on training workloads, AS9100D change records, CMMC configuration baseline — that governs every model in production.
WHAT WE'RE LOOKING FOR
• 7+ years of hands-on ML engineering — not slide-deck AI, not OpenAI API wrappers • Hands-on experience fine-tuning large language models in production • Designed and run evaluation frameworks • Debugged a production model that was doing something wrong, found root cause, and shipped the fix
You are comfortable with:
• QLoRA, LoRA, PEFT — hands-on fine-tuning on real GPU hardware • HuggingFace ecosystem: Transformers, TGI, PEFT, Datasets • LLM evaluation design: benchmark selection, human eval protocols, automated eval pipelines, calibration against ground truth • Retrieval-augmented generation quality: embedding precision, FAISS index management, context compression • Kubernetes — you will be deploying into K3s • U.S. Person status — training workloads touch ITAR-controlled technical data
Exceptional candidates also bring:
• Government/defense AI experience — ITAR/CUI data handling, CMMC-scoped systems, or classified compute • Familiarity with formal verification (Coq, Lean, Isabelle) — our proof layer produces machine-checkable certificates • Active Secret clearance, or willingness and ability to obtain one • On-premises GPU cluster operations: CUDA tuning, NVIDIA GPU Operator, MicroK8s, model serving optimization • Aerospace, defense, or manufacturing AI background
WHY THIS ROLE MATTERS
You will own models that matter. The intelligence platform you calibrate directly informs whether we pursue a $2M government contract. The CAD LLM you improve helps engineers build avionics instruments that go into aircraft. The retrieval quality you maintain shapes how every autonomous agent in our fleet reasons.
You will work on problems most AI engineers never touch. Formal verification of AI explanations. ITAR-compliant on-premises fine-tuning. Sovereign AI infrastructure built from first principles. A government intelligence platform with provenance hashing and Merkle proofs on every claimed fact.
You will operate at genuine scope. Director-level. Own a Line of Effort. Hire your own team. Interface directly with the executive layer. Your work matters from day one.
You will be part of something rare. A defense contractor building an AI-first operating system with formal governance, sovereign infrastructure, and an autonomous agent fleet that is already operational. This is not a roadmap. It exists.