AI Engineering

From AI prototypes to production systems.

I help teams adopt AI and put it to work: from the first prototype to systems that hold up in production, and the team workflows that make it stick. Startup founder (SpeedInvest · Expa · Antler) and CTO for 10+ years. I take the work from strategy all the way to implementation.

/Prototype & build

AI prototypes built quickly to prove the value, then hardened for production (agentic systems, orchestration, evals and guardrails).

/Integrate

Connect AI to your platform: MCP servers, Claude and ChatGPT apps, plugins and extensions, API and tool integrations.

/Adopt & enable

AI adoption strategy and hands-on training for your team. AI coding, agentic development and new ways of working.

/Knowledge & workflows

Knowledge bases and automated workflows that put your company's data and processes to work.

Over two decades building production software as an engineer, architect, CTO and founder, now focused on applied AI. I work with teams on scoped, project-based engagements, whether they're starting out with AI or scaling what they already have.

Founder at OpenTide.ai and Head of AI Engineering at OLBG. Previously founder at Genie. Creator of Genie Framework. Postgraduate diploma in AI & Deep Learning, UPC Barcelona. Master's degree in Advanced Computer Science, University of Liverpool.

Agentic systemsLLM orchestrationContext engineering MCPAI codingClaude PyTorchPythonTypeScript Deep learningAI strategy

My work

What I'm building, for my clients and the open-source community.

Products
Tide · Agent: every change comes with its why and approachTide · Diagram: a live call graph of the running appTide · Preview: the running app next to its live request logsTide · Verification: every changeset checked before it counts as doneTide · Docs & code: generated documentation next to the sourceTide · Search & Git: find anything, then commit and pushTide · Tests: the app’s test suite, run inside the kernelTide · Audits: accessibility, code quality, performance, security, SEOTide · Tools: 38 built-in tools the agent works withTide · REPL: evaluate code inside the running kernelTide · Settings: a model for each job: agent, planning, import, verification

The IDE reinvented for AI. Agent-first by design: plan with acceptance criteria, build against the live running app, and every change is verified before it commits. Built-in tests and audits, on a formally specified kernel.

DenoTypeScriptMCPClaude
Semnas · LibrarySemnas · Brief: one summary at a timeSemnas · Reader with AI summary and audioSemnas · Key points in the marginSemnas · Ask about the articleSemnas · Reading view

Read-later app built around AI. Every article you save gets a summary, key points and an audio version you can listen to, and you can ask it questions. A recommender learns what you read and fills your Discover feed and daily Brief.

LLMsRecommenderText-to-speech
YouKaraoke · Seven stems separated from one song, with sections detectedYouKaraoke · Current output: stem model, lyrics transcription, chart and tempo/key analysis

YouKaraoke

AI music analysis and stem extraction on self-hosted models: analyses songs from any audio source, splits them into instrument and vocal stems, detects tempo, key and song sections, and transcribes the lyrics.

Audio MLDemucsWhisperSelf-hosted models
Open source

The highly productive web framework for Julia. An ecosystem of 15+ open-source packages spanning servers, reactive UIs, ORM, and deployment. Grown to thousands of users across academia and industry.

JuliaWebOSS

The web framework for AI agents. Agents build and change a running app through gated MCP tools. Each change is a typed mutation, verified before it commits and kept in an append-only log you can roll back. Token-efficient by design: agents work on precise app entities instead of reading entire codebases.

DenoEvent-sourcingFormal methods
Genie Agents: an agent works through specs, BDD, plan, code and validation to build a data app

Genie Agents

Agentic system that generates Genie applications from custom-trained models. EU grant-funded, and my on-ramp from framework tooling into AI engineering.

AgentsCodegenFine-tuning

Genie Builder

Visual UI builder with a low-code Julia backend and a built-in AI assistant, so developers ship data apps without the boilerplate.

Low-codeAI assistantJulia
AI at OLBG
Worktron · Projects: tasks, pull requests, and the latest AI audits with their costWorktron · Audits per project, run on demand or hooked into GitHubWorktron · Fleet-wide audit definitions, such as the SEO competitive audit

Worktron

Agent-orchestration platform running autonomous agents in production. Agents audit code for security and SEO and run competitive analysis, triggered on pull requests or on a schedule. Model-independent: works with Anthropic, OpenAI, Kimi, DeepSeek and more.

AgentsOrchestrationMulti-model
OLBG.brain · Ask it about itself: a shared team memory for notes, decisions and skillsOLBG.brain · Admin: skill promotions waiting on a team decider, and procedures Claude noticedOLBG.brain · Installed as a Claude extension, with per-tool permissions

OLBG.brain

AI-driven knowledge management for teams, as a Claude extension. Versioned notes, decisions and skills with role-based sharing; Claude turns repeated procedures into skills teams can promote, and flags leaked credentials and duplicates.

Knowledge managementAccess controlMCPClaude
OLBG AI Search: an AI answer above the regular results for “football tips”, with live tips, odds and links

OLBG AI Search

AI answers at the top of OLBG’s site search: live tips with odds, tip counts and links to the matching pages, shown above the regular results. Powered by RAG over the site’s content, local classification models and hosted LLM APIs.

RAGAI searchClassification
OLBG Bot · Ask in plain language, like “best tip for tomorrow”OLBG Bot · Answers come back as live tip cards, with odds and tipster consensusOLBG Bot · Tipster profiles with profit history and strike rate

OLBG Bot

The OLBG Tips Bot: ask in plain language for tips, tipsters or odds, and get answers built from the site’s live data, returned as rich tip and tipster cards. Combines retrieval, in-house classifiers and AI APIs to understand each question.

ChatbotRAGLLMs
Client work
Nextmol · Installed as a Claude extension, with 30 platform tools and per-tool permissionsNextmol · Claude connected to the platform through its MCP server

Nextmol

MCP server and Claude plugin that connect a computational-chemistry platform and its API to Claude, so researchers can drive the platform directly from AI.

MCPPythonFastMCPClaude
Neuroblox: a corticostriatal circuit for Parkinson’s, with drug, stimulus and deep-brain-stimulation inputs and live signal outputs

Neuroblox

The interactive UI for a neural-circuit simulation platform, used by neuroscientists to model brain interventions in silico.

JuliaNeuroscienceUI
Quantum Savory Studio: a quantum network laid out on a map, with link distance and delay

Quantum Savory Studio

Visual studio for designing and simulating quantum networks on the QuantumSavory toolkit: place nodes on a map, configure slots and protocols, and run simulations.

JuliaQuantum computingUI

Work experience

2025–now

Founder

Building the next generation of AI-native development tools.

AgentsMCPTypeScriptDeno
2025–now

Head of AI Engineering

OLBG, UK sports-media & betting affiliate platform

Leading AI adoption and engineering across a high-traffic, regulated product.

AgentsOrchestrationKnowledge systemsPyTorchPython
2019–2025

Co-founder

Genie, Amsterdam / Barcelona

Founded and led a venture-backed developer-tooling company around the Genie Framework.

JuliaTypeScriptCodegenFundraising
2013–2019

CTO

OLBG, Bristol, UK

Built and led the product and engineering team behind the company’s platform.

Team buildingProductArchitecture
2011–2013

CTO / Founder

Aprefis (Washington DC / Miami) & Advidi (Amsterdam)

Built a cruise-booking price optimisation platform (Aprefis) and a high-traffic ad-serving and affiliate marketing platform (Advidi).

Price optimisationAd techRuby

Earlier: Team lead and web developer roles across Bristol and Bucharest (2000–2011), building and leading web product teams in PHP and JavaScript.

Writing & speaking

Books

Talks & community

  • 2017–24JuliaConSpeaker / organizer: Berkeley, London, Baltimore, Eindhoven
  • 2016–25Barcelona Julia Users GroupOrganizer
  • 2015–17IronHack BarcelonaMentor & trainer

Education

  • 2025–26Postgraduate Diploma, AI & Deep LearningUniversitat Politècnica de Catalunya (UPC), Barcelona
  • 2018–20MSc Advanced Computer ScienceUniversity of Liverpool
  • 2016, 18Certified Scrum MasterScrum Alliance
  • 2007–09MSc Computing (Cybersecurity)University of Bucharest

Let's talk

Trying to get AI from prototype to production, or want a senior pair of hands who can architect and build?

Show email address