99%
AI cloud infrastructure cost reduction
I reduced AI cloud infrastructure spend from about $25K/month to about $300/month with ephemeral Railway infrastructure.
Nürnberg, GermanyEU Blue Card holderFully authorized to work in GermanyNo visa sponsorship required
AI engineer
Deploying AIwhere workactually happens.
I build as an Forward deployed engineerAI engineerfor AI teams that need production systems, not slideware.
I turn ambiguous customer problems into deployed AI and full-stack systems, then codify the learnings into reusable engineering patterns.
Customer ambiguity to deployed AI systems
deployment-loop.session
1/scope customer outcome
2/build customer outcome
3/retrieve customer outcome
4/evaluate customer outcome
5/deploy customer outcome
This portfolio is also the proof: a CV-aware assistant, pgvector retrieval, Railway backend, Vercel frontend, and spec-driven delivery.
Evidence, not adjectives
I open with outcomes recruiters remember.
I designed this page like a product launch: a simple message, bold numbers, and interactive proof that I understand production AI systems, not just frontend polish.
5x+
Team output increase
I used AI-assisted, spec-driven full-stack delivery to reduce handoffs and increase team output. The autonomous pipeline turns specs into implementation, QA-style Playwright MCP checks, fixes, and regression tests for production-ready apps.
90%+
Automated test coverage
I build with spec-first regression coverage so production systems stay safe to change.
5+ yrs
Production-grade systems
I have shipped full-stack systems across AI tooling, fintech, healthcare, and automation.
Interactive AI proof
A portfolio that behaves like an AI system.
Ask George's CV answers questions about my background using public-safe CV and project data retrieved from pgvector before the model responds. The RAG/eval walkthrough shows how retrieval, grounding, and evaluation work together to keep AI answers useful and honest.
Ask George's CV
LLM chat · RAG tool access · grounded answers
Hi - ask me about my AI systems, FDE fit, full-stack delivery, cloud/backend work, projects, or development style.
Tool traceCollapse
- retrieve_portfolio_contextWaiting for a recruiter question before querying public-safe CV/project chunks.
Grounded RAG + Eval Loop
A visual demo of reliable LLM deployment
Recruiter query: “Can I build production LLM systems?”
01
Intent
Classify the recruiter question and decide whether CV retrieval is needed.
02
Retrieve
Search embedded CV/project chunks using vector + metadata signals.
03
Rerank
Prioritize evidence that directly supports the answer.
04
Generate
Answer only from retrieved context and avoid unsupported claims.
05
Evaluate
Score retrieval relevance, citation coverage, risk, and helpfulness.
I built multi-provider AI agent infrastructure with LLM orchestration, RAG pipelines, pgvector retrieval, and ephemeral Railway services.
similarity 0.94 · source: CVI reduced AI infrastructure spend from about $25K/month to about $300/month while using an autonomous spec-to-QA pipeline to help increase team output 5x+.
similarity 0.91 · source: CVRetrieval relevance94%
Citation coverage88%
Unsupported claims0
Metrics are illustrative demo metrics for explaining reliability patterns, not production claims.
Career narratives
From full-stack delivery to AI deployment leadership.
My career has moved from full-stack delivery into AI systems through customer-facing ownership, production rollouts, and measurable business impact.
2025 - Present
Lead AI Engineer · Autonomous Agent Infrastructure
I build AI agent infrastructure with multi-provider LLM orchestration, RAG pipelines, pgvector retrieval, and ephemeral Railway execution environments.
2022 - Present
Lead Software Engineer · Customer-Facing Delivery
I own end-to-end full-stack delivery across healthcare, fintech, automation, and internal AI tooling, with a focus on reliability, cost, and speed.
2021 - 2022
Backend Engineer · Compliance-Critical Systems
I built authentication, CI/CD, and frontend architecture for compliance-sensitive financial systems where correctness and maintainability mattered.
Selected launches
Projects that prove the positioning.
These two projects show how I think about product, architecture, and delivery: one public AI portfolio system, and one private automation build with sensitive implementation details kept private.
Portfolio meta-project
This AI/FDE Portfolio Website
I built this public portfolio as a working AI system: CV-aware chat, Railway Express inference, PostgreSQL pgvector retrieval, Voyage embeddings, OpenRouter LLM inference, a Vercel-hosted Next.js frontend, a Grounded RAG + Eval Loop demo, and Spec Kit/spec-driven development.
Next.jsRailwaypgvectorOpenRouterVoyage
Private AI automation
Forex & Cryptocurrency Trading Agent
An in-progress private trading and automation agent where I am exploring market monitoring, strategy orchestration, risk controls, backtesting/evaluation, and agentic workflows. It is not financial advice and makes no return, profit, or live-performance claims.
Private build - architecture details shared selectively.
Private buildAutomationRisk controlsEvaluation