Information, Not Chaos: Complexity Without Complication.
High Entropy LLC works on the systems-level architecture, security, and operational reliability of production ML and AI.
Our focus is the infrastructure that makes machine learning and AI systems work reliably in production – and increasingly, the security and validation architecture those systems require as they take on more autonomous and consequential roles. Most engagements involve some combination of these concerns, though the balance varies.
The shared underpinning is twenty-five years of designing distributed systems for hard, high-scale, high-stakes problems: ad ranking and click prediction, search infrastructure, malware detection, biometrics and forensics for adversarial inputs, AR hardware integration, fault-tolerant distributed workflow, and AI agent infrastructure.
Production ML and AI systems architecture
How do you build the infrastructure that makes machine learning and AI work reliably at scale, under real production constraints?
This is the work we’ve done across our careers and where most engagements live. The shape varies by client, but the work generally falls into:
- Architecture review and recommendation for ML/AI systems that need to scale, evolve, or be made more reliable
- Design work for new platforms or major rearchitectures
- Hands-on implementation for the hardest 20% of a build, where the standard playbook doesn’t apply
- Senior advisory for teams building production AI capabilities for the first time, including the engineering practices and team structure that make those capabilities sustainable
Past contexts: Google ads ranking and AdSense, Google search quality, Google TTS, Google Maps regional data, Meta ads experimentation, Meta privacy and data misuse prevention, Meta AR, DARPA media forensics (MediFor), DARPA logistics (LogX), US Army biometrics search at scale.
Security and validation for AI systems
A subset of our work focuses specifically on the security architecture and validation infrastructure that AI and agent-based systems require. Standard application-security frameworks don’t cover agentic identity, prompt-injection-resistant tool use, model-output filtering, evaluation methodology, or governance structures appropriate to AI capabilities. We help teams figure out what good looks like for their context.
Engagements in this area are typically narrower in scope than systems-architecture work: targeted reviews, threat modeling, evaluation framework design, and governance work informed by NIST AI RMF and related standards.
Past contexts: Atlassian (Senior Security Architect, AI Trust and Safety Lead – built the AI security program from the ground up), Johns Hopkins Information Security Institute (Cloud Security course design and instruction), public work in agent infrastructure (Engram), and ongoing work with AI safety organizations.
Who we are
High Entropy LLC is a small senior-only practice based in Maryland.
The practice is led by Christopher Monson, PhD – a 25-year veteran of distributed systems, ML platforms, and security architecture, with senior engineering and architecture roles at Google, Meta, Atlassian, and DoD programs at Data Machines Corp. PhD in machine learning and optimization from Brigham Young University. Author of Introduction to Programming for the Independent Student and co-author of Practical Cryptography in Python (Apress, with Seth Nielson). Three issued US patents in adversarial software detection. Current and former lecturer at Johns Hopkins Information Security Institute.
We work with a small bench of senior collaborators – practitioners with deep, specific expertise in adjacent areas – whom we bring into engagements requiring complementary depth. Each collaborator has been vetted through prior working relationships; we don’t subcontract to people we haven’t worked with closely before. The bench is intentionally small and senior-only.
How we work
Senior people doing senior work. We don’t pyramid; the person scoping the engagement is the person delivering it (or, when an engagement requires multiple specialties, working alongside a similarly senior collaborator). We don’t have juniors to staff or interns to manage. This keeps the work consistent, the rates honest, and the engagement model simple.
Designed for what comes after. We don’t just solve the hard part of a problem. We set up the scaffolding, code structure, team composition, and operational practices that make the rest of the work tractable for the people who’ll do it after we leave. An engagement that produces a brilliant artifact but a confused team isn’t a successful engagement.
Substance over packaging. Our deliverables are written for the people who have to act on them, not for executives who just want a summary. We aim to make the implicit explicit and the complex tractable, not to produce decks that look good but don’t help.
Selectivity over scale. We take on a small number of engagements at a time and decline work that isn’t a good fit for our expertise or the client’s circumstances. This serves both sides: we deliver better work when we’re not stretched, and clients get our actual attention rather than our calendar’s leftovers.
Engagement inquiries
For new engagements, technical questions, or general inquiries: chris@highentropy.com.