Principal Engineer · AI-Accelerated Systems

I modernize the systems businesses run on - at AI speed.

Hi, I'm Jeremy - a Principal Engineer pairing systems architecture with AI-accelerated delivery: production RAG systems and high-velocity modernization for Starbucks, T-Mobile, and F5.

See my work Download résumé

North StarStarbucks go-forward data model

12+ML models benchmarked

6 mo → 1 wkdeploy readiness

Available now

Where would you like to start?

Four quick doors in - take whichever fits why you're here.

See my work

Five recent builds - a production RAG system, a live analytics platform, and more.

Browse projects

Where I've been

Amazon, Starbucks, T-Mobile, F5 - and what I led at each one.

View timeline

Get to know me

How I work, my approach to applied AI, and what I value.

Read about me

Let's work together

Open to full-time roles now, and freelance engagements too.

Get in touch

About

A builder who makes complex systems calmer.

I move fluidly between Python, Java/Spring Boot, C#, Node.js, and Go - choosing the right tool for the problem instead of forcing every solution into one stack. That flexibility has paid off across e-commerce, telecom, and cloud infrastructure.

My sweet spot is modernization: refactoring legacy monoliths into microservices, migrating on-prem systems to cloud-native architectures, and weaving in modern observability and deployment practices - without the downtime that usually comes with it.

These days I build AI into production systems: RAG with ChromaDB, sentence-transformers and MMR-tuned retrieval, and AI-driven analysis of legacy APIs during migration. Parity between old and new validated against both mock data and real production data pulled into a test system. Old-school rigor, new-school speed.

On the AI side the work is orchestration, not model training: LangChain and LangGraph, retrieval tuning, and AI-assisted migration. The projects below are where it runs.

Selected work

Things I've built recently.

A mix of production platforms and developer tooling. Open any card for the full case study.

● Live · wa-creel.jeremyveleber.com

WA Creel

Turns WDFW creel reports and NOAA current data into actionable fishing insights for Puget Sound anglers - with real statistical rigor under the hood.

PythonSQLiteGCP Cloud RunLeafletChart.js

Read the case study

wa-creel.jeremyveleber.com

WA Creel - Puget Sound creel survey dashboard

wa-creel.jeremyveleber.com

Problem

WDFW creel reports and NOAA current predictions exist as raw tabular data. Anglers had no way to turn that into real decisions about where and when to fish.

Solution

Multi-metric analysis - CPUE, Poisson-based success rate, and a composite Hotness Score.
Significance testing - expands date windows until sample sizes are reliable.
Trip Planner wizard - weighted recommendations by species, area, and timespan.
NOAA integration - slack-current windows via multi-station averaging.

Stack

Python with a custom HTTP server, SQLite + GCS persistence, interactive visualization with Leaflet + Chart.js, Dockerized on GCP Cloud Run with automated CI/CD via Cloud Build. Refactored from a 2,464-line monolith into single-responsibility ~300-line modules.

Outcome

A production analytics platform serving recreational anglers, built on 13 years of survey data - where statistical methods guarantee recommendations meet minimum confidence thresholds and respect current regulations.

Visit the live site

Production RAG · Open source

claude-code-search

A production RAG system that lets Claude Code retrieve exactly the right code with minimal token usage.

PythonChromaDBsentence-transformersMMRRAG

Read the case study

Problem

Claude Code burns tokens reading whole codebases to find where functionality lives - broad scans with no guarantees.

Solution

A ChromaDB vector store chunked by class and function, with retrieval tuned by Maximal Marginal Relevance (MMR) to balance semantic relevance against diversity - so Claude never reads redundant or irrelevant code. A file watcher keeps the index live on every change.

Stack

ChromaDB persistent index, sentence-transformers embeddings benchmarked across 12+ HuggingFace models (semantic accuracy, inference speed, memory footprint), Python, CLI/API integration for Claude Code workflows.

Outcome

Open source, and a major win in token efficiency. LangChain and LangGraph orchestration runs the assistant on this site today.

View on GitHub

Open source · GitHub

api-gateway-hub

A Backend-for-Frontend API aggregator with Redis caching and resilient retry logic.

PythonFastAPIRedisPostgreSQL

Read the case study

Problem

Frontends calling many external APIs create latency bottlenecks and expose keys - with no caching, retries, or rate-limit protection.

Solution

A BFF service aggregating three external APIs with intelligent Redis caching, exponential-backoff retry, and stale-cache fallback on failure.

Stack

FastAPI 0.115 async, Redis 7.2 with tiered TTLs, httpx + tenacity for retries, PostgreSQL 16 for request logging.

Outcome

One unified endpoint, a 95%+ cache hit rate via cache-aside with stale fallback, and 84% test coverage using testcontainers.

View on GitHub

Open source · GitHub

authkit

A production-ready authentication microservice with JWT and OAuth2 social login.

TypeScriptNode.jsPostgreSQLOAuth2

Read the case study

Problem

Hand-rolled auth invites vulnerabilities; managed services like Auth0 add cost and vendor lock-in.

Solution

A self-hosted service with a dual-token pattern (15-min JWT + 7-day refresh), OAuth2 social login via Passport.js, and rate-limited endpoints - a drop-in Auth0/Clerk replacement.

Stack

Node 20 + Express 5 in strict-mode TypeScript, bcrypt + Zod validation, express-rate-limit, PostgreSQL 16 with cascade integrity.

Outcome

100+ tests over 70% coverage with real-DB testcontainers, Docker-Compose ready, and no auth vendor costs.

View on GitHub

Open source · GitHub

eventstream-api

A real-time event aggregator streaming GitHub, HackerNews, and Reddit via Server-Sent Events.

PythonFastAPIPostgreSQLSSE

Read the case study

Problem

Real-time dashboards mean polling many APIs, normalizing different schemas, and pushing updates - direct client polling is slow and duplicative.

Solution

Background polling of three external APIs, a unified event schema, and SSE streaming to connected clients, with PostgreSQL deduplication by external ID.

Stack

FastAPI 0.115 full-async (SQLAlchemy + httpx), SSE with a broadcast connection manager, PostgreSQL 16, Docker-Compose with k8s manifests.

Outcome

A streaming API over 70% coverage with per-source health checks, rate limiting, and optional API-key auth - ready to scale behind a load balancer.

View on GitHub

Experience

Five chapters.

Feb 2025 - Jun 2026

Latest

Sirrus7

Principal Engineer / Modernization Architect

Led platform modernization for T-Mobile, Starbucks, and an educational game platform - cutting deployment readiness from six months to under a week with a standardized Kubernetes + Istio platform, and developer onboarding from weeks to days.

Aug 2024 - Feb 2025

Ascendion / Starbucks

Senior Manager / Lead Software Design Engineer

Designed a unified data integration platform consolidating product data, and led modernization of transaction processing with zero-downtime migration strategies.

Nov 2018 - Apr 2024

Nortal

Lead Software Design Engineer

Modernized enterprise systems into hybrid legacy/modern operational models, migrating hundreds of APIs to modern frameworks with zero downtime.

2015 - 2018

F5 Networks

Senior Java Developer

Redesigned state-machine software for improved responsiveness, built advanced content-caching systems, and strengthened CI/CD pipelines.

2010 - 2014

Amazon

Senior Software Design Engineer

Designed a scalable photo pipeline, and built BI platforms on Kinesis, Redshift, and EMR.

Toolkit

What I reach for.

AI & ML Engineering

RAG systemsChromaDBMMR retrievalPrompt engineering

Languages & Backend

PythonJava / Spring BootGo

Cloud & Infra

AWSKubernetesIstio

Data & DevOps

MongoDBKafkaCI/CD

Open to roles & freelance

Let's build something solid.

Hiring for a senior or principal backend role - or have a modernization problem that needs a steady hand? I'd love to hear about it.

Emailresume@jeremyveleber.com LinkedInin/jeremyveleber GitHubgithub.com/jjveleber RésuméDownload PDF (2026)