I design AI agent systems that coordinate, learn, and ship real work. Grounded in 15+ years of hardware infrastructure, HPC, and platform engineering at scale — from 700+ compute nodes to production agent teams that operate like trained employees.
I'm building Agent MA — an AI agent platform where agents have layered personas, measurable competency levels, and defined roles. They coordinate as a team, learn from every interaction, and get better over time. The platform is built by the same agent team it's designed to sell — proof of concept and product in one.
At Tenstorrent, I pioneered Agentic AI before it was a category — building MCP-based agents that cut support ticket volume by 70% and saved teams hours of manual work daily. I managed 700+ compute nodes and 400+ AI accelerators, automated provisioning for 600+ servers, and transitioned a reactive IT organization into a proactive Platform Engineering team serving 800+ engineers.
Before that, I was a Data Architect at Jabil — building near real-time analytics pipelines for global factory operations, converting monolithic systems to microservices, and leading zero-data-loss migrations. That foundation in data architecture, hardware infrastructure, and operational automation shapes every AI system I build today.
Since going independent, I've shipped over twenty production systems in three months — from AI agent platforms to federated search engines to bare-metal orchestration. I own every layer from the server rack to the user interface.
Building coordinated multi-agent systems with layered personas, role-based access, and competency levels that improve with use.
20+ production systems in three months — agent platforms, search engines, infrastructure tooling. Production-grade is the only grade.
Vault-managed secrets, encrypted document storage, certificate-based auth, Fernet encryption, zero-trust networking. Security is architecture, not an afterthought.
Led engineering teams at every stop, from 8-person squads to 800-engineer platforms. Now architecting AI agent teams with the same rigor.
From hardware infrastructure and HPC to AI agent architecture — always at the intersection of systems design and execution.
Designing and building Agent MA — an AI agent platform with layered persona assembly, competency-based skill levels, role-enforced access control, and coordinated multi-agent teams. Running production infrastructure on bare metal: Kubernetes, Helm, Harbor, Vault, Prometheus. 20+ systems shipped in three months.
Pioneered Agentic AI solutions that cut support volume 70% — before it was a category. Managed 700+ compute nodes and 400+ AI accelerators across two datacenters. Led platform engineering for 800+ engineers, a Tiger Team of 20+ resolving critical hardware failures, and transitioned the organization from reactive IT to proactive Platform Engineering.
Managed DevOps, SRE, and SaaS operations across AWS, Azure, on-prem, edge compute, and IoT. Architected CI/CD pipelines spanning iOS, Android, Linux, and NUC deployments. Maintained and scaled DataStax, Spark, and Cassandra clusters.
Transformed a distributed team of 8 engineers across the US and overseas. Architected end-to-end CI/CD including multi-cloud and air-gapped deployments. Defined the pipeline roadmap for Kubernetes and legacy VM environments.
Built near real-time analytics pipelines for global factory operations using Kubernetes, Kafka, Elasticsearch, and Ansible. Converted a monolithic test framework to microservices. Led a zero-data-loss platform migration across 8 teams. Saved the business millions through automated reporting and manufacturing metrics.
Software architecture and infrastructure engineering at scale.
Since leaving Tenstorrent, I've been shipping — applying everything I've learned about systems design, security, and AI to build tools I believe in. Each one started with a real problem and was built with the same rigor I'd bring to enterprise infrastructure.
A platform for building AI agent teams that coordinate, learn, and get better over time. Agents talk to each other across users, share context in group conversations, and operate within role-enforced boundaries. Bring your own Claude subscription, manage your API keys, and invite others through your MA. Kubernetes-native, mobile-friendly, and self-healing by design.
Learn More →A federated search engine across 25+ marketplaces — designed as a plugin architecture where each provider is an isolated adapter with its own Fernet-encrypted config. AI classifies queries into intent, a 4-layer relevance scoring engine ranks results in real-time via SSE streaming, and Prometheus + OpenTelemetry trace every request from ingestion to render. One query, dozens of sources, full observability.
A multi-tenant ML inference platform with namespace-isolated deployments, HPA-driven GPU scaling, and scoped RBAC. JWT + API key auth, security headers middleware, rate limiting per tenant, S3 model storage, Stripe billing. Users deploy models through a simple API — the security and scaling happen automatically.
Musicians shouldn't need to be in the same room. A low-latency networked MIDI system in C, speaking raw UDP with RTT measurement, cross-platform between Windows and Linux. Plug in your keyboard, connect to a session, and play together.
Insurance sales management built on AES-256-GCM document encryption, brute-force protection, and role-based access (admin / manager / seller). Concurrent bulk import via worker pools, Claude-powered PDF and spreadsheet extraction, multi-vendor quote comparison. Every customer document is encrypted at rest — not optional, not configurable.
Privacy-first job matching — no credentials stored, no data retained. Claude tailors resumes per role, sentence-transformers score semantic fit, Playwright automates applications. Multi-service architecture with Celery task queues, each service scoped to a single responsibility. Your data stays yours.
Plus: SRE infrastructure agents, Rust file-sync CLI, GPU-accelerated background removal, document scanning pipeline, LLM caching proxy, self-hosted Excalidraw, Docker Compose to Helm converter, Ansible playbooks for Dell PowerEdge servers, bare-metal orchestration via MAAS, and more.
Every system I build starts with architecture — the right boundaries, the right failure modes, the right agent roles. I've designed platforms spanning 700+ compute nodes across datacenters, and now I architect AI agent systems on bare metal: Kubernetes, Helm, Harbor, Vault, Prometheus. I think in systems, not features.
Rust when correctness and performance matter. Go for concurrent services. Python for rapid iteration and ML. C when you need to talk directly to hardware. TypeScript for rich UIs. I don't have a favorite language — I have favorite outcomes.
At Tenstorrent, I built MCP agents that cut support tickets 70%. Now I'm building coordinated multi-agent teams with layered personas, competency tracking, and role enforcement. AI that operates like a trained team — not a chatbot bolted onto a dashboard.
Security isn't a checklist you run at the end — it's how you design the system from day one. Vault for secrets management, certificate-based authentication, encrypted storage at rest, zero-trust networking, automated credential rotation. I've managed infrastructure serving 800+ engineers where a breach would halt product development. That shapes how I build everything.
If your organization needs AI architecture, agent systems design, or someone who can build intelligent automation on real infrastructure — I'd like to hear about it. 15+ years of hands-on experience from hardware to AI, at companies like Tenstorrent, Jabil, Spirent, and HP. Let's talk.