Deep Orbit Labs
Agents · MCP · Connected hardware

AI agents and connected devices, built end to end.

I build AI agents and MCP servers that companies can run in production, and the hardware, firmware and apps underneath them when there's a device involved.

Start a conversation See what I build

What I build

Fixed scope where possible, so you know what you're getting and when.

An MCP server for your API

Your existing API exposed as a remote MCP server, so Claude, ChatGPT and other agents can use your product. Tool design, OAuth, rate limits, tests, docs, deployed.

Typically 2 to 4 weeks

Agent security review

A review of your agent or MCP deployment: who the agent acts as, credentials, tool permissions, admin surfaces and prompt injection boundaries. Written findings, plus the fixes.

Typically 1 week

Agent evaluation harness

An evaluation set and scoring pipeline for your agent or classifier, with synthetic ground truth where no labelled data exists, and a cost versus accuracy comparison across models.

Typically 2 weeks

Connected hardware

From sensor to app: circuit boards, firmware, Bluetooth, the companion app, and the agent layer on top. Prototypes built to be manufactured, not just demoed.

Scoped per project

Selected work

All public. Click through and check.

  • Registry as an MCP server
    Built an MCP server that lets any agent discover other agents through IAB Tech Lab's Agent Registry. TypeScript, Google ADK.
  • Agent credential hardening
    Required operator credentials on the admin surface of an open source seller agent, across REST and MCP. Followed by key management for the buyer agent.
  • Embedding model evaluation
    Benchmarked embedding models on a ~700 category taxonomy using 704 synthetic test pages generated with Claude. Best result 75.0% top-1, 94.5% top-5, and found the smallest vector size that kept that accuracy.
  • Bluetooth load-cell gauge
    A complete connected instrument: 4-layer circuit board, nRF52840 Bluetooth, 24-bit load-cell sensor with calibration and a written error budget, firmware, an iOS app reading it live, and a steel and printed enclosure.
  • Agent standards
    Led IAB Tech Lab's agentic advertising program: the agent registry (onboarding Google, Meta, Amazon and TikTok), the reference buyer and seller agents, and the Agentic Real Time Framework.
  • Open source contributions
    Shipped code into Prebid.js, OpenRTB and the AdCP protocol through public review.

About

Deep Orbit Labs is Miguel Morales. Most recently I led the agentic standards program at IAB Tech Lab, where I built the agent registry as an MCP server and fixed credential handling across the reference buyer and seller agents.

Before that I was co-founder and CTO at Lucidity, an advertising analytics platform, and VP of Engineering at Infillion, where I was the first employee. I design hardware too, from circuit boards to firmware to the app.

  • Based in Los Angeles, working with clients remotely
  • Stack TypeScript, Python, Java and Scala, MCP, A2A, Google ADK
  • Hardware KiCad, nRF52 and Bluetooth LE, OpenSCAD
  • Elsewhere GitHub · LinkedIn

Have something to build?

Tell me what you're working on. I'll reply within one business day.

miguel@deeporbitlabs.com