Toby Peterson

Principal Forward Deployed Engineer

I'm a software engineer. I came up through full-stack web development and these days most of what I build is enterprise AI — the governed, auditable kind. The first part is what makes the second part work.

Toby Peterson

About

Principal engineer with 7+ years shipping and maintaining production software, now working where large language models meet real business systems. I build governed AI platforms: Azure-native knowledge graphs and vector retrieval where authorization is enforced at the retrieval boundary rather than hoped for in the prompt, answer layers that cite their sources and return insufficiency instead of fabricating, and agent workflows wired into the platforms a company already runs on. My day-to-day spans the Anthropic/Claude API, Microsoft Foundry, and the Model Context Protocol — structured tool calling, context and token-budget management, prompt caching, and evaluation harnesses that prove a change actually improved output. The other half of the job is integration: joining Entra ID, Microsoft Graph, PostgreSQL and Cosmos pipelines, PSA, telephony, and accounting platforms into event-driven services with explicit permission boundaries. I came to engineering deliberately — a business degree, a full-stack immersive, then years of production work well before the current wave of AI enthusiasm. That foundation is exactly what separates a system people can depend on from a demo that falls over in month two. One principle I keep coming back to: arithmetic belongs in a SQL engine, not a language model, and every number a system reports should trace back to its source.

Lynnwood, WA

How I build

  1. Arithmetic belongs in a SQL engine, not a language model

    Every number a system reports should trace back to the file it came from, and a reader should be able to check it. A model that does its own math produces answers nobody can audit — and an answer nobody can audit is not an answer, it's a guess with good posture.

  2. Authorization belongs at the retrieval boundary, not in the prompt

    A system that asks a model to respect permissions has already lost. The only data a model cannot leak is data it was never handed. Enforce access where the records are fetched, and the prompt stops being a security surface.

  3. A system that can't say "not enough evidence" can't be trusted when it answers

    Returning insufficiency is a feature, not a gap in coverage. Confident fabrication is the failure mode that costs you the user — usually not the first time it happens, but the first time they catch it.

  4. Internal tooling is never more than one release from interrupting a thousand people

    I learned this maintaining a platform 16,000 people used daily. Change control, backward compatibility, and testing before shipping aren't process overhead on a system everyone depends on. They're the job.

Selected work

Experience

Principal Forward Deployed Engineer

centrexIT 2025 — Present
  • Principal engineer on an Azure-native knowledge graph and vector retrieval layer over Microsoft 365 and line-of-business data, where authorization is enforced at the retrieval boundary and every answer carries citations and an audit trail — driven from prototype to working multi-source pilot
  • Ship agent workflows and retrieval pipelines on the Anthropic/Claude API, Microsoft Foundry, and the Model Context Protocol — structured tool calling, context and token-budget management, prompt caching, and evaluation harnesses that prove a change improved output
  • Design answer layers that return insufficiency rather than fabricate when evidence is thin, and analyst surfaces where every figure traces back to its source
  • Join Entra ID, Microsoft Graph, PostgreSQL and Cosmos DB pipelines, PSA, telephony, and accounting platforms into event-driven services with explicit permission boundaries and supportable deployment patterns
  • Shape product concepts, data models, and workflows for internal platforms and client-facing assessment products, pairing data collection with explainable scoring and evidence review
  • Own promotion gates and pre-release verification, set engineering standards and developer tooling, and mentor engineers across concurrent builds

Software Developer II to Senior Software Developer

centrexIT 2022 — 2025
  • Built and maintained production Python and PowerShell services — webhook receivers, polling daemons, CLIs, and containerized workers — with persistent state, structured logging, and automated deploys
  • Developed custom connectors against REST and GraphQL APIs, handling OAuth2 and token refresh, pagination, rate limiting, retry and backoff, and idempotency so repeated events could not corrupt downstream records
  • Connected PSA and ticketing (HaloPSA), RMM, telephony (3CX, Twilio), Microsoft 365 and Microsoft Graph, Entra ID, CRM, and billing systems (QuickBooks, Bill.com, Pax8) into coherent cross-system workflows
  • Modeled and populated CMDB records, mapped schemas between systems with incompatible data models, and ran bulk migrations of thousands of records with field-level reconciliation
  • Built event-driven flows for ticket lifecycle, alert triage, data synchronization, billing and reporting, and zero-touch incident remediation, including real-time services consuming platform event streams and websockets
  • Delivered endpoint patching, compliance posture, software deployment, and identity provisioning at scale, orchestrated on Rewst, Power Automate, and Logic Apps where they fit

Software Developer II

Expeditors 2019 — 2022

Promoted from Associate Software Developer

  • Became sole developer supporting the enterprise ITSM platform — roughly 16,000 users across more than 200 offices worldwide, with one engineer accountable for what all of them relied on
  • Designed and built applications end to end — front end, back end, data model, and deployment — setting priorities with stakeholders, making the architectural calls, and supporting them in production
  • Connected the ITSM platform to Active Directory and identity services, HR and asset data sources, and other line-of-business systems through APIs and scheduled data exchanges
  • Built across the service catalog, service desk tooling, and the process automation engine using PowerShell, JavaScript, SQL, and HTML/CSS — request forms, stored procedures, approval and routing rules, and ticket lifecycle automation
  • Operated under strict change control and backward compatibility on a system the whole company depended on daily

Full Stack Web Developer

Galvanize Inc 2018 — 2019
  • Deliberate career change into software engineering — six months full-time of daily pair programming, code review, and shipping working applications on deadline, covering JavaScript, React, Node.js, PostgreSQL, and RESTful API design

Clinician & Administrator

Prior Career — Behavioral Health 2010 — 2018
  • Spent 8 years in behavioral health leadership roles before transitioning into software engineering in 2018

Education

Certificate, Java Programming

University of Washington

Certificate, Full-Stack Web Development

Galvanize

Certificate, Social and Human Services

Seattle Central College

Bachelor of Science, Business Administration and Management

University of the Pacific — Eberhardt School of Business

Certifications

Rewst Clean Automation Certificate

Rewst Issued Aug 2025

Rewst Foundations Certificate

Rewst Issued Aug 2025

Certified SAFe 5 Practitioner

Scaled Agile, Inc. Oct 2020 · expired Sep 2021

Azure Fundamentals AZ-900

Microsoft Jan 2020 · expired Jan 2023

AWS Certified Cloud Practitioner

Amazon Jan 2019 · expired Jan 2021

CompTIA A+ Certified

CompTIA Jun 2018 · expired Jun 2021

Skills

Languages

  • Python
  • TypeScript
  • JavaScript
  • Java
  • PowerShell
  • SQL
  • HTML / CSS
  • Jinja

AI Engineering

  • Retrieval-Augmented Generation (RAG)
  • Vector Search
  • Knowledge Graphs
  • Agentic Tool Calling
  • Anthropic / Claude API
  • Microsoft Foundry
  • Model Context Protocol (MCP)
  • Context & Token-Budget Management
  • Prompt Caching
  • Evaluation Harnesses

Cloud & Platform

  • Azure Container Apps
  • Azure Functions (incl. Durable)
  • Azure Blob Storage
  • Cosmos DB
  • Azure Static Web Apps
  • Entra ID
  • AWS
  • Cloudflare Workers
  • Vercel

Backend & Data

  • Node.js
  • Express
  • PostgreSQL
  • Supabase
  • RESTful APIs
  • GraphQL
  • Webhooks / WebSockets
  • Event-Driven Services
  • Schema Design & Migration

Frontend

  • React
  • Redux
  • Next.js
  • Angular
  • React Native
  • Responsive Design

Systems Integration

  • Microsoft Graph API
  • Microsoft 365
  • SharePoint
  • Exchange Online
  • HaloPSA
  • 3CX
  • Twilio
  • QuickBooks
  • Bill.com
  • Pax8
  • Rewst
  • Power Automate

Engineering Practice

  • Git / GitHub Actions
  • CI/CD Pipelines
  • TDD
  • Agile / SAFe
  • Code Review & Mentorship
  • Linux / Unix

Contact

Reach out via email or connect on LinkedIn.