Mice Pápai

Full-Stack AI Engineer

Ten years building production software, based in Budapest. These days I focus on data- and ML-enabled products, mostly in Python and TypeScript.

I do my best work owning a feature end to end: training data, model pipelines, backend services, and the UI people use. Lately a lot of that is agent systems, orchestrators coordinating specialized subagents.

Most teams split that work across several specialists. I'm looking for a remote role on a senior team where that breadth is an advantage. And I build constantly on my own time: twenty years of side projects below.

Experience

Scale AI (2024–present)

Full-Stack AI Engineer
  • Own and run a data-annotation app end to end, across ML, backend, and frontend. In production it has labeled thousands of PDFs and hundreds of architectural drawings, the training data behind models and apps now live.
  • Wrote the CAD-file parsing and conversion libraries behind its annotation workflow.
  • Built a TypeScript viewer/editor library for architectural drawings. It started as the annotation surface and now multiple Scale apps reuse it.
  • Shipped LLM and applied-ML features for enterprise and government clients: market-research and due-diligence agents, and domain agents for architecture, fire safety, and legal work.
  • On two of those, built the multi-agent systems themselves: orchestrators coordinating specialized subagents, backed by pipelines for LLM document extraction and image classification.
  • Ship features across the core platform (FastAPI backend, React frontend).

Digital Thinkers (2023)

Tech Lead
  • Led the rewrite of a client's app onto a modern stack, working with a frontend developer, a backend developer, and designers.
  • Worked directly with the client, turning their goals into scope and roadmaps.
  • Set up the team's code-review process and quality standards, mentoring developers through reviews.
  • Streamlined the design-and-QA workflow across the team.

Vacuumlabs (2019–2022)

Node/React Full-Stack Developer
  • Built fintech apps: banking systems and desktop cryptocurrency wallets for IOHK, in TypeScript, React, and Web3.js.
  • Led the frontend team, coordinating remote delivery with designers and backend developers across time zones.
  • Mentored peers on functional-programming patterns in TypeScript.

RoadRecord (2018–2019)

Python Backend Developer
  • Built backend services using Django, Redis, PostgreSQL, and RabbitMQ.
  • Shipped features in lockstep with the Angular frontend team.
  • Upgraded the codebase from Python 2 to Python 3 without disruption.

Endless Flux Labs (2018)

PHP/JS Full-Stack Developer
  • Built the company's publishing platform, a large WordPress application (PHP, MySQL, JS) that was the shared feature set every site was built on.
  • Used it to ship and run multiple custom Hungarian news sites.
  • Built advanced image-management features into the platform.

Freelancing (2015–2018)

Freelance Developer
  • Built dozens of web apps and scripts for clients on freelancer.com, picking projects that forced me to learn something new: scrapers, small ML and computer-vision projects, custom tools in Python, Java, PHP, and R.
  • Managed client relationships from initial contact to delivery.

Personal (2005–present)

Highlighted projects
  • Built cosmetic.science end to end, solo: an evidence-based skincare-ingredient reference covering 27,000+ ingredients. A Python extraction-and-synthesis pipeline pulls studies from PubMed and ClinicalTrials.gov and scores the evidence; the frontend is Astro, React, and TypeScript.
  • Validated the pipeline against a 900-case golden eval set, with property-based tests over the scoring math.
  • Build my own LLM and ML tools: a GRPO reinforcement-learning fine-tune for code generation, with rewards that run candidate code in a sandbox; a multi-model council that merges several frontier models into one confidence-scored answer; and a from-scratch agentic loop with multi-agent orchestration.
  • Built a cost-optimized pipeline that turns messy academic PDFs into clean Markdown.
  • Built earlier AI projects in 2010–2013: Csinszka, a Hungarian chatbot; computer-vision prototypes on neural networks; and Mixnose, a YouTube music-video recommendation engine.
  • Wrote social-graph analysis tools for Last.fm and Instagram.
  • Shipped Android apps (including a gyroscope-based social camera) and Drupal art-community sites.
  • Always building little tools for myself, like album_mosaic, a virtual vinyl collection for listening to whole albums (something Spotify makes hard).

Skills

  • Model training: fine-tuning, eval harnesses, and regression tests
  • Applied ML & GenAI: LLM features, RAG, agents
  • Data pipelines: turning raw data into something a model can actually learn from
  • Backends: production APIs in Python (FastAPI) and Node
  • Frontends: React and TypeScript, design included
  • Mentoring: code review and helping other developers level up

Languages

  • Magyar (native)
  • English (fluent)
  • Español (intermediate)
  • Русский (basic)

Interests

  • Photography
  • Music
  • Fine arts
  • Language learning
  • Neuroscience

Contact