This page no longer reads like a startup app gallery. It leads with production-scale payment systems, introduces three hardware-backed hero builds, and archives work that does not strengthen the principal systems and AI architecture narrative.
production proof
1
hero builds queued
3
commodity demos archived
2
Featured systems
PRODUCTIONEnterprise production system
The Home Depot
Payment Card Tender System
Contributed to the tender path that carries credit, debit, and gift card traffic across a nationwide store footprint and ecommerce, with observability and migration work on the critical path.
Constraint
Peak retail traffic and a zero-downtime mandate leave no room for opaque dependencies, weak telemetry, or brittle release procedures.
Trade-off
Favor observability-first design and buffered distributed flows over simpler synchronous coupling, even when that increases platform complexity up front.
Decision
Worked within a Go and GKE payment environment to improve telemetry, support zero-downtime tender migrations, and make high-risk paths visible during production incidents.
Outcome
Helped keep a platinum-tier payments domain measurable under load, with leadership-facing observability and sub-50ms p99 performance on critical paths.
ESP32 environmental nodes feeding a Raspberry Pi 5 gateway, local vision workloads, and a public GCP analytics board for backyard growth telemetry.
Constraint
Owning sensors is not the same as operating a system. The goal is a resilient edge network with useful telemetry, not a pile of disconnected boards.
Trade-off
Keep sensing, buffering, and first-pass inference local for latency and resilience; ship curated events and long-horizon aggregates to GCP for analytics and a public dashboard.
Decision
Use ESP32 nodes for capture, MQTT into a Pi 5 gateway for edge rules and Frigate, then normalize events in Cloud Run and persist time-series analytics in BigQuery.
Outcome
Turns a backyard project into proof of silicon-to-satellite thinking: hardware, enclosures, observability, edge AI, and cloud analytics in one system.
ESP32Raspberry Pi 5MQTTFrigateCloud RunBigQueryGrafanaTerraform
Next milestone
Bring the first outdoor node online and publish a live telemetry board.
BUILDING NOWHero build | self-healing edge ops
Independent build
Edge Fleet Control Plane
A control plane for Pi 5 and ESP32 hardware that manages heartbeats, signed OTA rollouts, incident timelines, and public health telemetry from one operator surface.
Constraint
A shelf of boards is inventory, not architecture. The missing piece is coordinated fleet health, rollout discipline, and observable failure handling.
Trade-off
Use a lightweight edge gateway and event bus instead of heavyweight cluster tooling so the system stays cheap, understandable, and deployable on real hobby hardware.
Decision
Run fleet coordination and local buffering on the Pi 5, push signed rollout metadata and incident state through Cloud Run APIs, and store historical reliability signals in BigQuery.
Outcome
Demonstrates that the same operational patterns used in enterprise systems can govern small edge fleets with real SLOs, rollback paths, and health visibility.
Raspberry Pi 5ESP32TypeScriptCloud RunBigQueryOpenTelemetryTerraform
Next milestone
Stand up signed firmware manifests and device heartbeat reporting.
BUILDING NOWHero build | machine vision + manufacturing
Independent build
Autonomous Print Cell
A Bambu P1S production cell with ESP32 power and environmental sensing, Pi 5 vision analysis, and cloud-side quality analytics for failed-print detection.
Constraint
A printer is a tool. A print cell becomes a systems project only when failures, energy use, and quality signals are captured and acted on automatically.
Trade-off
Run vision and anomaly checks locally so detection remains fast and inexpensive; send structured events to GCP for quality trends, auditability, and remote monitoring.
Decision
Pair the printer with ESP32 telemetry, local camera analysis on the Pi 5, and a Cloud Run ingestion pipeline that records print events, defects, and energy data into BigQuery.
Outcome
Creates a hardware-backed AI operations case study that blends manufacturing telemetry, edge inference, and cloud analytics without becoming another generic web app.
Bambu P1SESP32Raspberry Pi 5FrigateCloud RunBigQueryLooker Studio
Next milestone
Instrument the printer job stream and calibrate local failed-print detection.
Supporting systems
LIVELive supporting system
gimenez.dev Operator Console
The portfolio itself as a live operator surface: recruiter-facing RAG, admin telemetry, trace-linked logs, and a war room that documents how the system behaves in production.