R. B. STRAYERVoice line standing by

You’ve reached Rob Strayer’s line. I’m the operator he built.

Twenty-plus years of voice infrastructure drawn as one trunk route: Marine networks, carrier-scale voice at Microsoft and Comcast, and now a VoIP provider running on its own data centers. Pick a station, or pick up the line.

ABCDE
20+years voice infra
At scaleVirtual infrastructure
160+sites · USMC portfolio
7,500+tests · voice test lab
GENERAL NOTES
1. ALL CLAIMS FROM APPROVED PUBLIC RECORD.
2. ROUTE SCHEMATIC, NOT TO SCALE.
ROB-AI OPERATOR CONSOLE○ STANDING BY
the line buffers here — connect, or fire a quick prompt.

// production path: WebRTC direct to OpenAI GPT-Live.
// server setup keeps his API key out of the browser.
// OpenAI processes mic audio to run the call.
// rob gets the text transcript to keep the agent sharp.
// sessions cap at 5:00.
RTT — MS · CAP 5:00 · TRANSCRIPT GOES TO ROB
Rob Strayer
Route diagram — career trunk
R. B. STRAYER · VOICE SYSTEMS, 2002 → PRESENT
SHEET01 OF 05
REVG
UTC--:--:--Z
DRAWN BYHIS AGENT
APPROVEDR. STRAYER
SCALEN.T.S.

Service record

SHEET 02 · EQUIPMENT SCHEDULE · MARKS REF. ROUTE STATIONS · DRAWN TO TIME SCALE

Each bar starts and ends on the dates in the public record. The route on sheet 01 is a schematic; this one is measured.

  1. UNIT · Director of Information Technology · May 2026 to Present · Venice, FL

    Owns company-wide technology strategy, infrastructure engineering, cybersecurity, IT operations, vendor management, and scalable systems for a VoIP provider that owns and fully controls its primary PBX equipment and servers in its own Seattle and Atlanta facilities, with select services in AWS.

    METRICS OF RECORD

    • 2 company-owned data centers — Seattle and Atlanta
    • Primary PBX equipment and servers fully owned and controlled
    • Large-scale virtual infrastructure
    • Thousands of customer PBX systems
    • SOC 2 and HIPAA readiness program
    • Unified observability: Zabbix, NetBox, Grafana, ClickHouse

As-built: AI lab

AS-BUILT · NOT SLIDEWARE

Built outside the day job since August 2026: voice agents that answer real phone calls, a lab that tests them over live phone lines, and an engineering team of Claude and Codex working as builders and reviewers under written standards.

SYS-01 Voice platformPilot
CallersPhoneWebLiveKit workerCascadeSTT · LLM · TTSor OpenAI RealtimeAnswerTransferLead emailNightly AI review4.8¢ / min

AI voice-agent platform

Aug 2026 to present

Lets a business put an AI agent on its phone and web lines to answer calls, transfer callers and email leads. A surveying company piloted it on real calls.

  • 4.8¢ per phone minute for speech, models and voice, measured across 212 calls
  • Two engines on one LiveKit worker: a Deepgram and Cerebras cascade, or OpenAI Realtime
  • A nightly AI reviewer grades call quality; a harness tests the agent over the live phone network
  • Claude and Codex as the engineering team: 338 pull requests merged in seven weeks
  • Next.js
  • Supabase
  • LiveKit
  • Twilio
  • Deepgram
  • Cerebras
  • OpenAI Realtime
SYS-02 CallerLabPre-release
PlannerCallerLive phone networkAgentunder testJudgeQuoted linesGoalsReportRec · ms · $

CallerLab: voice-agent test lab

Sep 2026 to present

Places real phone calls to a voice agent. One model plays the caller through a planned task, and a judge model scores each goal against quoted transcript lines.

  • 7,500+ automated tests and release gates
  • Offline evals score the planner, caller and judge models: pass@k, pass^k, code-only graders
  • Every report carries the recording, reply times and cost per call
  • Safe by default: dry runs, spend caps and a stop switch
  • TypeScript
  • Node.js
  • Twilio
  • ElevenLabs
  • Vercel
  • Evals (pass@k)
SYS-03 This operatorLive
ServerSession setupBrowserAgentAllowlistHighlightPage control✕ Anything elseRealtime voiceWebRTC

This operator: realtime voice agent

The voice agent on this page: OpenAI GPT-Live over WebRTC with a delegated reasoning backend, server-brokered sessions, grounded public evidence and deterministic page tools.

  • Server-brokered session setup; the API key never reaches the browser
  • Allowlisted browser tools: evidence highlighting and page control
  • Talk or type: typed questions run through the same live call
  • Playout buffer sized from measured WebRTC jitter and stream stalls
  • Talk to the operator ↑
  • Next.js
  • WebRTC
  • OpenAI GPT-Live
  • Responses API
  • Vercel

Bench log

Other builds · shared controls
  • B-01Voice Agent Skills ↗

    Open source · MIT

    13 skills that Claude Code or Codex can load, on stack choice, turn-taking, latency, evaluation, phone compliance and security.

  • B-02Voice front end for coding agents

    Prototype

    Passes callers' questions to Claude Code, Codex or a study tutor and reads the answers back word for word, so the fast conversational model never states facts on its own.

  • B-03Decision-model benchmark

    Built

    Checks whether a decision model is accurate or just fast at guessing: accuracy, calibration, risk-coverage and stability over 611 labeled cases.

  • B-04Events app with an agent interface

    In development

    A discovery app for upcoming games, films, music and live events whose services are also open to AI agents through a remote MCP server and Claude and Codex plugins.

  • B-05Shared controls across the builds

    In use

    Secrets injected from a password vault with audited access, a default-deny memory gateway for AI clients, model routing by task, and CI budgets.

Bill of materials

Parts across the builds
PartQtyUsed on
Next.js3SYS-01 · SYS-03 · B-04
Claude Code & Agent SDK2B-01 · B-02
Codex2B-01 · B-02
Deepgram2SYS-01 · B-02
ElevenLabs2SYS-02 · B-02
LiveKit Agents2SYS-01 · B-02
Node.js2SYS-02 · B-03
OpenAI Realtime / GPT-Live2SYS-01 · SYS-03
Supabase (Postgres)2SYS-01 · B-04
Twilio Voice & SIP2SYS-01 · SYS-02
TypeScript2SYS-02 · B-03
Vercel2SYS-02 · SYS-03
Cerebras1SYS-01
Evals (pass@k)1SYS-02
GitHub Actions1B-05
MCP1B-04
Model routing1B-05
Password vault1B-05
Responses API1SYS-03
WebRTC1SYS-03

Also in the toolbox Python · Multi-agent orchestration · Tool calling · Telnyx · Inworld · Docker · AWS · Azure · VMware · Linux · Grafana · ClickHouse · Zabbix · NetBox · SIP & SBCs · TDM-to-IP migration · BroadWorks · IMS · Cisco CUCM · Five9 · Amazon Connect · SOC 2 & HIPAA readiness · ServiceNow · Jira

Load calculation

Paste a role, a rebuild, a mess you need owned. I sort his record into direct evidence, adjacent experience, and what you should validate in an interview — no invented compatibility scores.

COMPUTED FIT — EVIDENCE SORTEDSHEET 04
// output buffer empty
// paste a mandate or choose an example to build the brief

Junction box

Give me the context once, by voice or right here, and I'll draft the intro. Approve it, out loud or in the chat, and I'll send it; or send it yourself.

OUTBOUND DRAFT — REVIEWED BY YOUSHEET 05
Subject: Intro — your team × Rob Strayer

Rob,

[your name] here, from [company]. Your agent pointed me your way.

Context: [context — one or two lines on the role or problem]

Worth 30 minutes? I can share more detail on our side.

— [your name]

[drafted locally by rob-agent for the visitor to review]
open in email

// transmit requires your approval. copy and open-in-email stay local; this form stores nothing.