Skip to main content
$ls -la ./projects

Selected Work

PRODUCTION

Payment Card Tender System

The Challenge

Black Friday traffic spikes (order-of-magnitude load) with zero margin for error. Legacy monoliths failing under load with inadequate observability.

The Solution

Designed a distributed buffering architecture using Pub/Sub & Cloud Run to decouple ingestion. Built OpenTelemetry tracing pipelines to make the entire payment path observable. Led zero-downtime GCT migration to type-safe Go microservices.

The Impact

Zero downtime during peak volume (high-throughput daily transaction volumes). Sub-50ms p99 latency. 20% latency reduction from observability-driven optimization.

GoGCP Cloud RunPub/SubCockroachDBRedisOpenTelemetryPrometheusGrafana

arch:Event-Driven Distributed Systems

gcp:Cloud Run, Pub/Sub, BigQuery, Cloud Build

The Home Depot

Payment Card Tender System

scaleEnterprise-scale
availabilityPlatinum-tier SLO
latencySub-50ms p99
improvementObservability-driven

gimenez.dev AI Portfolio

The Challenge

Static portfolios fail to answer specific questions. Cloud LLM APIs are expensive at scale and leak context to third parties.

The Solution

Engineered a full RAG pipeline: local knowledge base, optional Supabase pgvector for semantic search, self-hosted inference via Inferencia (OpenAI-compatible). Deployed on GCP Cloud Run with Terraform IaC.

The Impact

Live production system demonstrating RAG architecture, edge-first AI thinking, and GCP deployment. Operational cost under $2/month.

Next.js 16TypeScriptAI SDKSupabase pgvectorTerraformCloud RunDocker

arch:RAG + Self-Hosted LLM on Cloud Run

gcp:Cloud Run, Artifact Registry, Secret Manager, Terraform

Personal Project

gimenez.dev AI Portfolio

cost~$2/mo
infraTerraform IaC
aiLocal RAG
deploymentCloud Run

Churnistic

The Challenge

Manual ML retraining was a bottleneck, causing model degradation and stale predictions that cost revenue.

The Solution

Automated the entire pipeline using TensorFlow.js & Firebase Functions with data drift detection triggers, validation gates, and automatic deployment of superior models.

The Impact

Improved model accuracy by 15%. Eliminated 5+ hours/week of manual engineering toil. Caught a data drift issue that manual processes missed for weeks.

TypeScriptTensorFlow.jsFirebase FunctionsReact

arch:Event-Driven ML Pipeline

gcp:Firebase Functions, Cloud Firestore

Personal Project

Churnistic

accuracy+15%
saved5h/week
triggerDrift Detection

Other Work

Real-Time Trading Engine

WebSocket-driven trading platform with Go backend, real-time market data streaming, and portfolio analytics.

GoWebSocketsgRPCPostgreSQL
Source