AI/ML Engineering Consultant — Local LLM Deployment & Security (Fractional, Ongoing)

We're a Montana-based commercial lender (Renegade Capital) that has built an in-house AI credit memo pipeline using open-source models running entirely on local hardware.

No cloud AI — all data stays on-premise for regulatory and data sovereignty reasons.

We need a fractional AI/ML engineering consultant (3-5 hrs/week) to serve as our technical point person on the model infrastructure.

This is an ongoing advisory + hands-on role.

What you'd do: - Own the technical direction of our local LLM pipeline (currently Qwen 3.5 27B via Ollama on dedicated hardware) - Help with prompt engineering, output validation, and pipeline improvements - Advise on model security, inference hardening, and data isolation architecture - Be the credible technical voice when we speak with SOC 2 auditors, bank partners, and investors about our AI infrastructure - Evaluate model upgrades, fine-tuning strategies, and quantization tradeoffs as the open-source landscape evolves You should have: - Real experience deploying open-source LLMs locally (Ollama, vLLM, llama.cpp, or similar — not just API wrappers) - Understanding of model quantization, context windows, and inference optimization on consumer/prosumer hardware - Ability to articulate AI architecture decisions to non-technical stakeholders (auditors, bankers, investors) - Python proficiency (our pipeline is FastAPI + async) - US-based (required — we need someone who can join calls with regulated financial partners in US business hours) Nice to have: - Experience with financial services, lending, or regulated industries - SOC 2 or compliance background as it relates to AI/ML systems - Fine-tuning experience (LoRA/QLoRA) on domain-specific data Engagement: - 3-5 hours/week, ongoing - Async collaboration + a weekly 30-min sync - Mountain Time preferred but any US timezone works To apply, please answer this in your cover letter: "Describe a model you've deployed locally (not via API).

What base model, what hardware, what serving framework, and what was the use case?"