Test In Sight Consulting
AI Agents for Test, Lab & Industrial Infrastructure

AI agents that drive real lab hardware. Repeatably. In production.

Two decades automating physical test labs — RF and wireless, satellite communications, 4G/5G, industrial control, enterprise networking — and the last three years building the agent layer on top of them. I design MCP servers and multi-agent orchestration that let Claude, Gemini and other LLMs safely operate live instruments, with the test-engineering rigor that makes it hold up outside a demo.

Start a conversation See the platform
10production MCP servers
200+typed instrument tools
20+ yrsphysical-lab test automation
Fortune 500on-prem agent deployment delivered
Spirent TestCenter · CyberFlood · Velocity/iTestCalnex SNE-X · AtteroNetScout L1 opticalKeysight Velocity · spectrum analyzersArista EOS · Cisco IOSTokalabs SDLVIAVI TM500 · TeraVMPyVISA / SCPI
Services

What I do for test organizations

Engagements are typically fractional or project-based, on site or remote, and start with a fixed-scope pilot on your own hardware so you can judge results before committing further.

MCP Servers for Instruments

A production-grade Model Context Protocol server per instrument or platform, built from the vendor's REST or native API, OpenAPI spec, or SCPI command set.

  • Typed tools only — no raw CLI passthrough to hardware
  • Session lifecycle: setup → discover → reserve → start → observe → release
  • Reservation-gated mutation; read/observe tiers separated from mutating tiers
  • Python (FastMCP) or Node.js, one Docker container per server, behind a streamable-HTTP gateway

Agentic Lab Orchestration

Multi-agent systems that treat your lab as code: an orchestrator with per-server sub-agents, LLM-driven tool routing and response synthesis across Claude, Gemini and OpenAI models.

  • Test-as-a-Service workflows from a single prompt: configure, run, observe, report
  • Evaluation harnesses: preflight probes, deterministic tool walkthroughs as the CI gate, LLM-as-judge smoke tests
  • Execution tracing and failure-signature analysis that feed corrections back into later runs

Forward-Deployed Architecture & Training

Hands-on technical leadership embedded with your team — the same role I've played for carriers, NEMs and network-security vendors.

  • On-prem deployment, upgrades and hardening of agent platforms
  • Automation strategy, tool selection and lab-consolidation reviews
  • Workshops for test engineers: MCP, agent evaluation, Velocity/iTest, PyVISA/SCPI automation
Proof of Work

A production agentic-AI platform over a multi-vendor lab

As lead architect of a Mission Control AI-agent platform (2024–2026), I designed and shipped ten MCP servers exposing a live test & measurement lab to LLM clients, plus the orchestration, packaging and evaluation layers that keep them safe to use. IP for this AI tooling is contractor-retained.

Spirent CyberFlood

55+ tools across 13 test types — eMix, HTTP throughput, DDoS, breach assessment, ZTNA and more — generated and hardened from CyberFlood's OpenAPI specification and validated against a live controller.

Node.js13 test types

Spirent TestCenter

Two servers: REST-based and native TCL, covering ports, stream blocks, RFC 2544 sequencing, topology and live statistics. Paired with a Python agent framework using specialized sub-agents and a self-healing execution layer.

Python / FastMCPREST + TCL

Calnex SNE-X

Network-emulator control — ports, impairment maps, templates, statistics — so an agent can stand up a delay/loss/jitter scenario, run a test through it and tear it down on request.

Python / FastMCPImpairments

L1 & L2 fabric

NetScout L1 optical switching plus Arista EOS (eAPI) and Cisco IOS L2 switches, giving agents the ability to build and tear down physical topologies between instruments and devices under test.

Lab-as-CodeTopology

Tokalabs SDL & Keysight Velocity

Lab-management and orchestration platforms exposed as tools, so reservations, resources and existing test assets are visible to the same agents that drive the instruments.

ReservationsLab management

Orchestration & evaluation

Chainlit orchestrator with per-server sub-agents over OpenRouter; Bun/TypeScript headless and TUI agents (CLI, library, HTTP+SSE); deterministic evaluation gates in CI; LangSmith tracing.

Multi-agentCI-gated

Ask for a live walkthrough against real instruments.

Results

Measured outcomes from twenty years of lab automation

The agent layer is new; the discipline behind it isn't. A few examples of what automation has delivered for previous clients and employers.

8 h/week → 24/7Tier-1 US carrier test teams moved from a scheduled manual window to continuous automated execution on Spirent Velocity/iTest/TestCenter
~⅔ CAPEX cutProjected RF-lab instrument savings for a Tier-1 NEM by pooling shared instruments over RF-over-fiber and optical switching
2 eng-weeks → overnightSatellite-modem regression cycle replaced by an unattended PyVISA/SCPI automated regression station
40 h → 4 hWi-Fi and HomePlug field-test measurement, one of ten completed automation projects with measured time reductions
How It Works

From instrument API to agent-ready

  1. Lab discovery. Inventory your instruments, APIs, topologies and the tests your team runs most; pick the first high-value target and define the safety boundaries.
  2. MCP server build. A tested, containerized server for that instrument, following the same lifecycle and packaging conventions used across the existing platform so later vendors plug in without changes to consuming agents.
  3. Agent workflow and evaluation gate. Wire the server into an orchestrator, prove an end-to-end scenario on your hardware, and put a deterministic walkthrough into CI so regressions are caught before an agent touches the lab.
  4. Hand-off and training. Your team receives the code, the evaluation suite and a workshop so they own it going forward.
About

Inti Sanchez

AI Solutions Architect and Forward Deployed AI Engineer for test, lab and industrial infrastructure. Founder of Test In Sight Consulting, a Florida LLC in continuous operation since 2020 and the vehicle for an embedded Spirent residency at a Tier-1 carrier and, most recently, the Mission Control agentic-AI platform engagement.

Before that: automation specialist for VIAVI's 5G RAN-to-Core program; pre-sales solutions engineer at Spirent advising NEMs, CSPs, hyperscalers and semiconductor customers; senior RF test automation engineer for satellite modems at ST Engineering iDirect; industrial-control robustness testing at GE Power; and seven years directing Wi-Fi and HomePlug validation automation through the Intellon → Atheros → Qualcomm acquisitions. Earlier work at Motorola, 7 Layers and Panasonic Mobile in 3G, Bluetooth and GSM test systems.

APNT USVIAVI SolutionsSpirent CommunicationsST Engineering iDirectGE PowerQualcomm AtherosMotorola

Toolbox & credentials

  • Applied AI: MCP · FastMCP · LangChain/LangGraph · Chainlit · Anthropic & OpenAI SDKs · OpenRouter · RAG (OpenSearch) · LLM evaluation
  • Lab orchestration: Test-as-a-Service · Lab-as-Code · L1 optical and L1–L3 switching · RF test & signal analysis · device reservation · CI/CD with on-prem runners
  • Platforms: Python · TypeScript/Node.js · TCL · C#/.NET · Ansible · Docker Compose · VMware/KVM · Wireshark custom dissectors
  • Education: M.S. Computer Engineering and B.S. Electrical Engineering, Florida International University · Post-Graduate Certificate, AI & ML Business Applications, McCombs (UT Austin) · Technology Leadership, Cornell
  • Base: Ocala, Florida — remote and on site across the US · English and Spanish
Contact

Tell me about your lab

Send a short note about the instruments you run and what you'd like an agent to do with them. I reply personally.