Cross-Platform
Engineer.
Mobile · Web · Agentic AI Engineering
16+ years shipping mobile products — React Native, Expo, iOS, Android. One codebase across all platforms: native apps, web, and PWAs from a single TypeScript source. Delivered in fintech, blockchain, media, and enterprise.
Full-stack capable — TypeScript, Node.js, REST APIs, CI/CD — and increasingly focused on AI Engineering: designing agentic workflows where specialized AI agents build, review, and verify real work under explicit boundaries.
Currently building two AI-native apps in production — StatsOS and DayOS.
AI Engineering
From LLM integration to agentic engineering systems.
Three years ago this was calling LLM APIs from mobile apps. It's grown into designing Agentic AI systems — where specialized agents research, build, review, and verify real work with defined responsibilities, not a single model doing everything unsupervised. Siteworks is the clearest evidence of that, below.
LLM integration
Connecting LLMs (Claude, GPT-4, Gemini) to products — chat interfaces, content generation, smart search, contextual suggestions.
AI-assisted engineering
Using AI to automate parts of the engineering process itself: PR reviews, release notes, CI/CD decisions, test generation.
Agentic workflows & orchestration
Designing systems where specialized AI agents build, review, and verify real work under explicit roles and boundaries — see the Siteworks case study below.
On-device & edge
Integrating on-device ML models and edge inference for features that need to work offline or with low latency.
Case Study
BG Siteworks — Agentic AI, in production.
Siteworks builds real business websites, but the product isn't the point — the engineering system behind it is. It's a multi-agent pipeline where a builder agent and an independent reviewer agent have separate responsibilities, and neither one is trusted blindly, including by me.
01
Research & verify
Public sources checked and recorded before a line of copy is written — no invented services, prices, or history.
02
Build & orchestrate
Claude Code researches, designs, and implements — then deploys and self-reviews its own output.
Claude Code
03
Independent review
A separate, read-only session evaluates the live build against sources, brand, UX, and technical standards.
OpenAI Codex
04
Verify & remediate
Every finding is independently accepted or rejected with evidence — not applied automatically — then fixed.
Claude Code
05
Fresh review & QA
A new reviewer session checks the fixes, hunts for regressions, and clears the build for deployment.
deployment-ready
Shared standards
Every project runs against the same written standards for design, content, SEO, and review behavior — not ad-hoc prompting per site.
Brand as constraint, design as freedom
Real logos and established brand colors are preserved as hard constraints. Everything else about the design is free to change.
The builder doesn't grade its own work
Codex reviews independently and read-only. Claude then evaluates each finding on its own merits rather than applying it blindly.
Execution boundaries
The pipeline runs autonomously through research, build, review, and fix — but stops before anything that touches a real domain or live integration without sign-off.
Claude Code and OpenAI Codex are the current implementation — the roles are what matter. A builder with deep context, an independent reviewer with none, and explicit boundaries around what either one is allowed to do on its own.
Automation
I build the tools I use.
If I find myself doing something more than twice, I write code for it. The result is a growing set of personal tools — bots, dashboards, systems — that handle the repetitive work so I can focus on the hard parts. These aren't side projects. They run every day. The same instinct, taken further, is what became Siteworks.
Invoice upload bot
view flow →Uploading receipts and invoices to solo.ro was a manual, multi-step process after every expense.
Telegram bot — send a photo or PDF, it uploads automatically via Playwright. Zero manual steps.
Think of something you do on a regular basis.
I probably have a way to automate it — or can build one.
Experience
16+ years in software engineering.
Client projects
·React Native Developer
NDAMultiple React Native client projects from scratch — fintech, retail, and enterprise. Expo EAS, white-label component systems, strict TypeScript throughout.
Tiexo
·Full Stack Engineer
Full-stack NFT aggregator on Solana — blockchain parsing, transaction building, marketplace REST API, and Next.js front end, owned end-to-end.
Moonlet
·React Native Developer
Non-custodial crypto wallet — React Native for iOS, Android, and Chrome Extension, plus a backend for remote screen composition.
Soundmix
·React Native Developer & Technical Lead
Live video streaming app for DJs — led mobile and backend team from recruiting and architecture to CI/CD and delivery.
Currently Building
Personal AI projects.
Side projects I'm actively building and using — both in public beta.
Health analytics from wearable data.
Processes Whoop exports and surfaces insights on recovery, sleep, and strain — with AI-generated summaries. Built with React, Expo (web), and LLM integration.
Skills & Stack