Daniel Raymond
I architect production-grade AI & the teams that ship it.
Fifteen-plus years across LLM systems, agentic AI, DevSecOps, and cloud. Currently leading enterprise AI at ATA LLC — before that, running my own consultancy and leading DevSecOps on a $500M defense program at Lockheed Martin.
The shortest path to a proof of concept is rarely the right path to a reliable system.
What follows are the systems I'm proudest of — production architecture, agent platforms, open source. Each one solved a real problem at scale.
- 01 Prose-First Agentic AI Org-wide rollout across engineering, QA, and product. +30% productivity; eliminated up to 100% of select manual work. Python · LLM · K8s 2026
- 02 Document Intelligence Platform LLM pipelines, retrieval, and structured extraction replacing fragile rule-based parsing. $250K+ annual efficiency gains. Python · AWS · LLM 2026
- 03 WebbDuck Self-hosted SDXL image studio: text-to-image, img2img, inpaint, smart extend, upscaling, LoRA management, searchable local gallery. Python · FastAPI · Diffusers 2026
- 04 DNADuck Face identity clustering and LoRA dataset prep for local image libraries. Persistent state, review/merge workflows, LoRA-ready export. Python · FastAPI · InsightFace 2026
- 05 Enterprise Automation Platform Lockheed Martin · $500M defense program. Eliminated two weeks of manual effort per release cycle. $1.2M annual savings. Terraform · K8s · CI/CD 2022–23
Architect for what comes after the demo.
Retrieval, orchestration, evaluation, infrastructure, security, approvals — the work that lets teams actually depend on the system.
Autonomy with kill switches.
Governance, auditability, and human approval where they matter — useful agents in real organizations, not lab toys.
Measurable behavior over vibes.
Regression checks, eval harnesses, observable workflows. If you can't measure it, you don't really ship it.
Local-first when it matters.
I work fluently across hosted APIs and self-hosted stacks — privacy, latency, and cost decide which one.
Tools I've put into real production — across hosted and self-hosted.
Frontier assistants where they earn their cost. Local runtimes where privacy or latency demands it. Glue that ages well.
Got a hard AI problem?
Production architecture. Agent platforms. Governance and evals. Build-vs-buy. I read every inbound — usually reply within a day.