AI Engineering
LLM-powered products from prototype to production: tool use, retrieval, evals and guardrails. I treat models as components with contracts, not magic.
- Claude & LLM APIs
- Tool use / MCP
- RAG & evals
- Prompt & context design
Managing Engineer at Nationale-Nederlanden. 15+ years building platforms for banking, pensions and hosting, and now leading teams where AI agents and engineers ship together through pipelines built to catch problems before release.
AI that works in production needs more than a model. It needs orchestration, and it needs a pipeline that won't let bad changes through.
LLM-powered products from prototype to production: tool use, retrieval, evals and guardrails. I treat models as components with contracts, not magic.
I have spent years recruiting, mentoring and leading scrum teams. Now I run coding agents the same way: clear tasks, scoped permissions, isolated workspaces and human checkpoints where judgement matters.
Pipelines that make shipping boring, even under DNB and EU regulation. Every change is built, tested, secured and provisioned as code before anyone hits merge.
This is the team from the mission above. Every card is live: dispatch a mission and watch who picks up what.
Sets direction, owns the trade-offs, makes the final call.
—
Turns a goal into scoped, parallel tasks with acceptance criteria.
—
Implements each task on its own branch, in a sandbox.
—
Reads every diff for bugs, security issues and simplicity.
—
Writes and runs the tests that prove the task is done.
—
Keeps READMEs, ADRs and changelogs true to the code.
—
Drives the pipeline: builds, previews, deploys and rollbacks.
—
Decides what is worth building and signs off releases.
—
Shapes the experience; reviews previews, not mock-ups.
—
Explores the preview like a real user, looking for surprises.
—
People own the goals, the trade-offs and the final call. Agents never merge to production on their own.
Research, scaffolding, refactors, tests and docs run in parallel — each agent in its own isolated branch.
Type checks, tests, previews and audits gate every change, whoever (or whatever) wrote it.
Every task, diff and decision leaves a trail: PRs, logs and reviews you can read later.
Every change to ghiata.com runs through GitHub Actions: built, type-checked, previewed on its own pages.dev URL, then deployed to Cloudflare's edge on merge.
on:
pull_request: # preview every PR
push: { branches: [main] } # ship main
jobs:
build:
steps:
- run: npm ci
- run: npm run check # types
- run: npm run build
deploy:
needs: build
uses: cloudflare/wrangler-action
with:
command: pages deploy dist Reviewers — human or agent — click a live URL instead of reading a diff blind.
Builds and type checks are required status checks. Red means it doesn't ship.
Static output served from Cloudflare's global network with zero servers to babysit.
Dependabot opens update PRs weekly; the same pipeline vets them.
Banking, insurance and hosting: software that faces regulators, auditors and millions of customers. Here's the history, the way I'd read it.
Leading engineering for the next-generation pension platform: people, architecture and delivery. Now bringing AI agents into how the team builds and ships.
Joined Nationale-Nederlanden as staff after three years with them as a consultant.
Built the platform behind every new NN pension product: Vue, Node.js, Lambda, Kinesis and RDS on AWS, all provisioned in CloudFormation. Grew a high-performing team through hiring, training and mentoring.
Credit-card lifecycle systems: customer-facing Angular + Spring, real-time Kafka streams and batch processing.
Created a cloud-agnostic, turn-key, white-label container orchestration solution on Kubernetes. Spent innovation Fridays building it and mentoring consultants.
Moved to the Netherlands. Built the software behind ABN AMRO's Corporate Banking website, including iDEAL integration and PSD2.
Help center, fraud-prevention and abuse-handling tools used daily by internal teams. Migrated 5,000+ help articles automatically.
BSc Informatics · MSc E-Commerce at the Bucharest University of Economic Studies.
Bringing AI agents into your engineering workflow, or tightening the pipeline they ship through? Let's talk.