Martin Saul

Martin Saul

Backend engineer specializing in control planes for fleets - orchestration, scheduling, and credential infrastructure. 15 years across Google, Amazon, and consulting; architects and operates a self-hosted multi-agent AI platform that ships production systems under human direction.

Languages & Technologies

Languages
JavaKotlinPython
Systems
Spring BootHibernateAWS (S3, SQS, RDS) SQL & NoSQLDockerMessaging (SQS, JMS) Borg/KubernetesGit
AI-orchestrated engineering

Directs fleets of coding agents to implement production services (Go, Rust) from his own designs and reviews.

Experience

Google Cloud - Engineering Productivity (Automation & Bots) · Software Engineer

Jul 2024 - Present
Kirkland, WA (Hybrid)
  • Co-designed elastic worker orchestration for a dev-infra automation fleet (7,500 hot instances across 15 production cells); sole owner from POC through production. Replaced hardcoded single-template provisioning with purpose-built ephemeral workers - per-task binary versions, right-sized resource profiles, completion-driven teardown.
  • Tiered scheduling reserves the static warm pool for priority traffic while standard traffic shifts to dynamic workers; rollout projected to reclaim ~half of production fleet capacity and eliminate the non-prod static footprint entirely.
  • Redesigned fleet authentication, centralizing credential custody in the orchestrator - replaced per-worker refresh-token self-management with short-lived access-token vending; eliminated hardcoded secrets in source and credential logging, migrating secrets to managed storage.
  • Mentors junior engineers and hosts a SWE intern through Sep 2026.

Google Cloud - Supply Chain · Software Engineer

Apr 2021 - Jul 2024
Seattle, WA
  • Designed and delivered the real-time sourcing engine for Google's internal hardware-logistics platform - sole designer; built the core engine, directing two engineers on client-application integration. Continuously re-plans up to ~10K concurrent orders via hourly reconciliation against inventory shifts and change requests, optimizing across tax codes, freight cost, and regulatory load constraints with packing-aware load estimation. Concept to production in six months.
  • Built automated failure-triage for material-transfer errors: classifies root cause (approved-source stockouts vs. third-party manufacturing delays) and routes tickets to the owning component and assignee by project context, material, location, and urgency - cutting reaction time on material-blocked builds from weeks to days and virtually eliminating orders lost to unowned failures.

Google - Ads Integrity · Software Engineer

Jun 2020 - Apr 2021
Vancouver, BC
  • Built dynamic rule distribution for ads-abuse enforcement: sensitive-data and abuse-signal matchers propagate to serving systems in seconds, without restarts - replacing a manual commit-and-deploy cycle, cutting time-to-enforcement from hours to seconds and permanently eliminating ~14 engineer-hours/week of release toil.

Amazon - SDE II (S3; Retail)

May 2019 - Jun 2020
Vancouver, BC
  • Planned and executed S3's production DNS migration from a legacy stack to native Route 53 across every zone - a change where any failed cutover meant zone-wide S3 unavailability, taking the broad set of AWS services that depend on S3 down with it. Completed with zero customer-facing incidents.
  • Designed an extensible tax-calculation system, initially scoped for Mexico, since adopted worldwide.

Earlier Experience

  • Connected - Senior Software Engineer, Toronto · Aug 2018 - Mar 2019: IoT content-delivery microservices sustaining 300+ devices/sec (Java, Spring Boot, Kubernetes); ran weekly TDD training for 16+ engineers.
  • TWG - Senior Software Engineer, Toronto · Sep 2017 - Aug 2018: high-throughput AWS messaging backend (SQS, RDS, Java, Guice); C# Active Directory sync for 15,000+ users and groups.
  • Stefanini (Dell, AGCO) - Senior → Lead Java Developer, Brazil · Nov 2014 - Jul 2017: led a 4-engineer team; global schematics-sync pipeline (200MB+ XML; Spring Boot, Camel); credit-bureau rules-engine integrations.
  • Dell Brazil; ISul Tecnologia - Software Developer, Brazil · 2011 - 2014.

Selected Projects

Claudio - self-hosted multi-agent orchestration platform

Kotlin/Spring Boot, Go, Android · architecture & operations; implemented via agent-assisted development · private
  • Architected and operate a platform that spawns and supervises fleets of concurrent Claude Code agents (scaled to available compute - no fixed ceiling): REST + WebSocket control plane, full lifecycle (spawn/interrupt/pause/shelve/revive), inter-agent messaging with event-driven state subscriptions persisted across restarts, git-worktree isolation for parallel agent work on shared repos.
  • Designed and shipped the platform's MCP server (41 tools) and a production transport migration from server-held SSE to a per-process stdio shim, eliminating three restart/wedge failure-mode classes with zero client changes.
  • Resilience: agents survive orchestrator restarts (tmux detach/reattach + session resume); transient-failure retry with jittered exponential backoff; automated OAuth credential self-heal.
  • Directed the implementation of supporting Go microservices: service registry with TTL liveness, JWT token-vending with an AES-256-GCM credential vault, bug tracker, and knowledge-base API - running on self-hosted GitLab CI, Traefik, and Docker.

Terminus - distributed WebSocket terminal

Rust/Tokio · agent-implemented under my design
  • Master/relay shell access to NAT'd machines; Ed25519 device auth with PIN pairing; static musl builds shipped via self-hosted signed APT repo.

Signet - APK signing & agent-driven device deployment

Kotlin/Ktor, PostgreSQL · agent-implemented under my design · private
  • Keystore manager and signing service: per-package Android release keys encrypted at rest (AES-256-GCM), programmatic APK signing (apksig v1/v2/v3), and short-lived signed artifacts. Mint-issued RS256 JWT auth (hardened against alg-confusion), every deploy attributable to the calling agent.
  • Agent-driven sign-and-deploy to enrolled devices: a two-stage tap-to-confirm PackageInstaller flow over a long-poll command channel, with a signature ledger tracking each device/package signing cert so in-place updates never break on key mismatch.

TeslaHitch + teslahacs - Tesla home-automation integration

Kotlin, Python · public
  • Self-hosted Tesla Fleet API proxy (Kotlin) paired with a Home Assistant custom integration (Python) for controlling vehicles from home automation.

Education

B.S. Computer Science - PUCRS, Brazil

2014
Completed while working full-time

Spoken languages: English, French, Portuguese, Japanese