GNOMUS / projects / résumé / work / about / log
connecting ···

Cloud / AI Platform Architect  ·  Federal / FedRAMP  ·  US Citizen, active Public Trust

Most agentic AI is a demo.
Mine runs in production.

I architected and shipped an enterprise agentic AI platform — chat, workflow orchestration, and sandboxed code execution — inside a FedRAMP-High-aligned federal health boundary. It was onboarded by the agency's security organization and adopted across multiple groups while most of the field is still piloting.

Read the résumé→ See what I've shipped→ [email protected]✎
role
cloud / ai platform architect
in production
agentic platform · fedramp-high aligned
based
EST · remote only
open to
principal / staff roles

the evidence

01

In production — not a pilot

An enterprise agentic AI platform running live inside a FedRAMP-High-aligned federal health boundary: conversational ops, workflow orchestration, and sandboxed code execution.

02

Governed for government

Azure AD on-behalf-of identity scoping, three-tier governance, immutable hash-chained audit, and air-gap-capable deployment — built to survive a security review, not just a demo.

03

The whole stack, shipped

~14 services, NDJSON streaming, a model-agnostic router (Anthropic · Bedrock · Vertex · Azure · Ollama), and CI/CD governed by 21 ADRs with SBOM + image signing on AKS.

04

Adopted, then expanding

Onboarded by the agency's security organization and adopted across multiple additional groups — security as the beachhead, not the obstacle.

and I don't just architect it — I ship it

Six products in rotation, built and running. Every one below is a real screenshot of the thing working.

Nº 01

AgenticWork

Agent orchestration for companies that have a compliance team. Model routing, sandboxed execution, on-demand tool synthesis, and an audit trail you can take to a board meeting.

  • 70+ node workflow engine — agents, approvals, code execution, integrations
  • Model-agnostic router: Anthropic, Bedrock, Vertex, Azure Foundry, Ollama
  • On-behalf-of identity scoping with immutable, hash-chained audit
Shipping visit
AgenticWork — product screenshot
AgenticWork — running in the terminal

Nº 02

OpenAgentics

The open-source edition. Same orchestrator, same router, same audit log — you host it, you own it, one command to install.

  • One-command install — Docker Compose stack with a guided setup wizard
  • 11 built-in MCP tool servers: AWS, Azure, GCP, Kubernetes, GitHub, o11y
  • Human-approval gate and append-only audit, self-hosted end to end
Shipping Apache-2.0 visit
OpenAgentics — product screenshot
OpenAgentics — running in the terminal

Nº 03

AgentiCode

An AI coding agent that lives in the terminal — 26 built-in tools, a VS Code extension, and a desktop shell. Built to work on real repositories, not toy examples.

  • Terminal-native agent loop with 26 built-in tools
  • VS Code extension + desktop app share one engine
  • 286 automated tests covering the tool surface
Shipping
AgentiCode — product screenshot

Nº 04

Synth

On-demand agent tooling. The agent describes the capability it needs, Synth writes a one-shot tool, and a human approves it before a single line runs. No static registry to maintain, nothing left behind.

  • Mandatory human approval — the reviewer sees code, scopes and intent
  • Credentials injected at sandbox boot; the model never sees your tokens
  • Ephemeral by design — nothing to enumerate, nothing to compromise after
Shipping visit
Synth — product screenshot
Synth — running in the terminal

Nº 05

Brainbow

Shared browser control for humans and agents — one Chromium, seen through pixels instead of DOM snapshots, with a cinematic recorder built in.

  • CDP screencast at ~30fps to a live WebSocket viewer
  • Human and agent drive the same session simultaneously
  • Declarative .tape scripts turn a run into a recording
Foundation MIT visit
Brainbow — product screenshot
Brainbow — running in the terminal

Nº 06

PatchBay

A studio cable-routing visualizer — drag-and-drop virtual patching of a real rig, with a validation engine that catches the wiring mistake before you make it.

  • Every device face generated from data — ~29 panels, one visual system
  • Validation rules as pure, unit-tested predicates over the patch graph
  • BFS signal-path tracing colors where the audio actually goes
Shipping
PatchBay — product screenshot

the platform, running

Not mockups — these are screens from the live system: the chat surface, a real 22-node orchestration graph, the admin control plane, the terminal coding agent, and multi-provider model routing.

A 22-node agent workflow open on the Flows canvas
orchestration · 22 nodes, 21 edges
The platform chat surface answering a question with a visible reasoning trace
conversational ops · streamed, with reasoning
Admin console showing token burn, active flows and MCP fleet health
control plane · 67 admin surfaces
The terminal coding agent running a shell tool call and showing a diff
sandboxed code execution · real tool calls
Provider management showing routing tiers, health and cost per provider
model routing · providers, tiers, cost
The flows gallery listing deployed workflows with run counts
the flow library · deployed and scheduled

what I'm looking for

  • Principal / Staff — AI or Cloud Platform Architect
  • Permanent remote only · EST · US Citizen, active Public Trust
  • Teams shipping agentic AI into regulated / federal production
The full résumé→ Start a conversation→