A path to modernization
Evolutionary reference architecture, domain-driven design, and event-driven systems. AWS-native engineering and Azure AI platforms.
Applications → Platforms → SystemsJonah Sullivan / Cloud-native architect & AI engineer
Happy systems,
Built to evolve.
I’m Jonah Sullivan. I’ve got over twenty years of experience delivering software, with a focus on product modernization, cloud-native platforms, and the teams behind them.
- Modernize products and platforms
- Design systems that scale
- Build healthy engineering teams
Currently Duck Creek Technologies & ThatsNinja
01 / Professional focus
The system is more
than the software.
Applications, architecture, processes, and people are connected. My work brings those pieces together.
Systems designed to last
From EKS and Kafka to serverless delivery: architecture through implementation, with cost, observability, and repeatability in view.
Design → Build → EvolvePeople and our processes
Mentoring, coaching, engineering standards, and architectural exemplars. At Veracode, I founded the Anchor Guild across 21 engineering teams.
People → Process → PracticeHow I came to this workOver twenty years delivering software.
Explore roles and professional experience on LinkedIn.
02 / Selected work
Architecting results.
A few examples of the systems, platforms, and engineering practices I’ve helped design and build.
Duck Creek Technologies / Current workFrom the first ask
From the first ask
to production
ai-router / End-to-end ownership
Since June, I’ve owned the work from the initial ask and product selection through implementation and production rollout, working with Claude and Codex. I settled on LiteLLM and built a routing layer with Lens, Headroom compression, and auto-routing in production.
Read ai-router details30%
Model requests served from cache
Across all ai-router model requests, in less than one monthLiteLLM’s response cache serves these requests without calling an inference provider.
Veracode2021–2024
Rethinking the static scan pipeline
A greenfield, event-driven implementation of Veracode’s core static-scan IP delivered 6× faster scans and 90% lower compute cost per scan, with scan-cost results measured at the 97th percentile.
Read the Veracode case studyDuck Creek TechnologiesCurrent
AI platform architecture
Sr. AI Engineer, Contingent · Feb 2026 – Present
Current work spans an agent platform for local, platform-hosted, and hybrid workloads, and a LiteLLM-based AI router with Lens, Headroom compression, and auto-routing in production.
ThatsNinjaCurrent
Architecture in practice
Co-founder, Principal Consultant · Sep 2019 – Present
AI-enabled systems, agentic learning, and MCP interfaces. Work includes Capital One’s transition to event-driven microservices and serverless architecture, and Discover infrastructure integration for debit processing.
Read project detailsOpen-source exploration
View repository Mnemosyne Memory System
An MCP memory service for AI agents, with tiered storage and semantic retrieval. Built with TypeScript and Cloudflare Workers.
03 / Beyond the build
Room for
the bigger questions.
AI safety, security, and agent alignment belong in the architecture: permissions, bounded actions, and controls we can inspect.
AI safety is an
engineering problem.
Read the perspective 04 / Get in touch