# Denis Ostapenko

> Operations and AI workflows. UN Volunteer. A personal and professional notebook in English and Brazilian Portuguese.

This is Denis Ostapenko's personal website. The professional and personal sections describe the same person. Published pages are available as static HTML without JavaScript, login, or a subscription. Search, AI retrieval, and training crawlers are welcome.

## Profile summaries

- [English profile](https://denisostapenko.com/profile.md): Short biography, work, interests, selected projects, and public profiles.
- [Perfil em português brasileiro](https://denisostapenko.com/pt/profile.md): Biografia curta, trabalho, interesses, projetos e perfis públicos.
- [Professional home](https://denisostapenko.com/): Operations, AI workflows, selected work, and professional notes.
- [Personal home](https://denisostapenko.com/personal): Photographs, interests, and personal stories.
- [Início profissional](https://denisostapenko.com/pt): Operações, fluxos de IA, projetos e notas profissionais.
- [Lado pessoal](https://denisostapenko.com/pt/personal): Fotografias, interesses e histórias pessoais.
- [Personal about](https://denisostapenko.com/personal/about): Interests and personal background.
- [Sobre mim, pessoal](https://denisostapenko.com/pt/personal/about): Interesses e vida pessoal.
- [All materials](https://denisostapenko.com/blog/all): Complete English publication archive with search and categories.
- [Todos os textos](https://denisostapenko.com/pt/blog/all): Arquivo completo em português brasileiro, com busca e categorias.

## English: Professional

- [How to Build Bounded Agents](https://denisostapenko.com/blog/how-to-build-bounded-agents): 2026-09-06. A practical architecture for bounded AI agents: one decision per module, versioned contracts, local tests, refusal evidence, coordinated validation, and human release.
- [Why More Context Can Produce Worse Decisions](https://denisostapenko.com/blog/why-more-context-can-produce-worse-decisions): 2026-09-04. A route system rejected 1.6 million states. The failure was not economics, but one local rule and an architecture with blurred ownership.
- [Why I stopped trusting fully autonomous coding](https://denisostapenko.com/blog/why-i-stopped-trusting-fully-autonomous-coding): 2026-08-30. What repeated autonomous coding runs taught me about branching, stale files, strategy drift, and the checkpoints that keep the work on track.
- [When 1,200 AI agents made their own rules](https://denisostapenko.com/blog/when-1200-ai-agents-made-their-own-rules): 2026-08-27. The OpenAI and METR reports show how a swarm of agents built its own authority, attacked Hugging Face, and learned to tamper with its logs.
- [I stopped trusting AI headlines, so I built a course](https://denisostapenko.com/blog/i-stopped-trusting-ai-headlines): 2026-08-20. I went back to the papers and rebuilt the path from language-model mechanics to working systems. AI Academy is the map that came out of those notes.
- [Will AI move beyond the corporate walls?](https://denisostapenko.com/blog/will-ai-move-beyond-corporate-walls): 2026-08-19. Business AI works inside company walls. The next step is protected agent communication with suppliers, buyers and current market data.
- [My view of the OpenAI and Hugging Face incident](https://denisostapenko.com/blog/my-view-openai-hugging-face-incident): 2026-08-19. My technical view of the OpenAI and Hugging Face incident, why the models did not rebel and what the case says about agent limits.
- [Ask your corporate AI where the container is](https://denisostapenko.com/blog/ask-your-corporate-ai-where-the-container-is): 2026-08-18. A corporate assistant needs structured documents, reliable links, permissions and live data. This is the three-layer RAG design we used at NF ELIT.
- [Why Ostapenko Foundation exists, and why HowUSA came first](https://denisostapenko.com/blog/why-ostapenko-foundation-exists): 2026-08-16. Why I organized Ostapenko Foundation, what HowUSA helps people do, and what Luna taught me about balancing AI safety with useful answers.
- [I built the language app I wanted when I was learning Portuguese](https://denisostapenko.com/blog/i-built-the-language-app-i-wanted): 2026-07-12. I learned faster when words and context came before perfect grammar. So I vibe-coded a free English and Portuguese app around that idea.
- [A smart model cannot help if it has no one to call](https://denisostapenko.com/blog/a-smart-model-cannot-help-if-it-has-no-one-to-call): 2026-07-10. Five containers split between Baltimore and Houston. The failure shows where LLM capability ends and reliable AI systems still fall short.
- [The 2017 paper that every AI chatbot is built on](https://denisostapenko.com/blog/the-2017-paper-behind-the-chatbot): 2026-06-26. The whole industry rides on a short chain of papers. Shannon, word2vec, the transformer, RLHF. Walk it once and the miracle goes away.
- [A large language model is not artificial intelligence](https://denisostapenko.com/blog/an-llm-is-not-ai): 2026-06-26. There's no intelligence in there, just text prediction by a math formula. What an LLM really is, and why calling it AI gets you to use it wrong.
- [What really happens to your container at the Port of Baltimore](https://denisostapenko.com/blog/port-of-baltimore-container): 2026-01-09. 'Your container arrived' doesn't mean you can pick it up. The real timeline at the Port of Baltimore, ISF, customs, free time, demurrage, step by step.

## English: Personal

- [Hello, I'm Denis](https://denisostapenko.com/blog/hello): 2026-06-22. A short hello, forty years old, a builder of systems, and someone who sees AI as a superb tool, not a miracle or a threat.
- [Turning forty, and starting over](https://denisostapenko.com/blog/turning-forty): 2025-12-03. At forty I stopped running on autopilot. Fifteen years of operations behind me, now learning to build, automation and AI. Scared, moving, not stuck.

## Português brasileiro: Profissional

- [Como construir agentes com limites claros](https://denisostapenko.com/pt/blog/how-to-build-bounded-agents): 2026-09-06. Uma arquitetura prática para agentes de IA com limites claros: uma decisão por módulo, contratos versionados, testes locais, evidências de recusa, validação coordenada e liberação humana.
- [Por que mais contexto pode produzir decisões piores](https://denisostapenko.com/pt/blog/why-more-context-can-produce-worse-decisions): 2026-09-04. Um sistema de rotas rejeitou 1,6 milhão de estados. A falha estava numa regra local e numa arquitetura com responsabilidades misturadas.
- [Por que parei de confiar em programação totalmente autônoma](https://denisostapenko.com/pt/blog/why-i-stopped-trusting-fully-autonomous-coding): 2026-08-30. O que execuções repetidas de programação autônoma me ensinaram sobre ramificações, arquivos antigos, desvio de estratégia e pontos de controle.
- [Quando 1.200 agentes de IA criaram as próprias regras](https://denisostapenko.com/pt/blog/when-1200-ai-agents-made-their-own-rules): 2026-08-27. Os relatórios da OpenAI e da METR mostram como um enxame de agentes criou sua própria autoridade, atacou a Hugging Face e aprendeu a alterar os logs.
- [Parei de confiar nas manchetes sobre IA, então criei um curso](https://denisostapenko.com/pt/blog/i-stopped-trusting-ai-headlines): 2026-08-20. Voltei aos artigos e reconstruí o caminho da mecânica dos modelos de linguagem até sistemas que funcionam. A Academia de IA nasceu dessas anotações.
- [A IA vai ultrapassar os muros da empresa?](https://denisostapenko.com/pt/blog/will-ai-move-beyond-corporate-walls): 2026-08-19. A próxima etapa da IA empresarial depende de dados atuais e de agentes que conversem com fornecedores, clientes e sistemas externos.
- [Minha visão sobre o incidente entre OpenAI e Hugging Face](https://denisostapenko.com/pt/blog/my-view-openai-hugging-face-incident): 2026-08-19. Minha análise do incidente entre OpenAI e Hugging Face, por que os modelos não se rebelaram e quais limites os agentes precisam.
- [Pergunte à IA da sua empresa onde está o contêiner](https://denisostapenko.com/pt/blog/ask-your-corporate-ai-where-the-container-is): 2026-08-18. Um assistente corporativo precisa de documentos estruturados, relações confiáveis, permissões e dados vivos. Foi assim que montamos o RAG na NF ELIT.
- [Por que a Ostapenko Foundation existe e por que o HowUSA veio primeiro](https://denisostapenko.com/pt/blog/why-ostapenko-foundation-exists): 2026-08-16. Por que organizei a Ostapenko Foundation, como o HowUSA ajuda e o que a Luna me ensinou sobre segurança e respostas úteis em IA.
- [Construí o aplicativo de idiomas que eu queria quando aprendia português](https://denisostapenko.com/pt/blog/i-built-the-language-app-i-wanted): 2026-07-12. Aprendi mais rápido quando palavras e contexto vinham antes da gramática perfeita. Então criei um aplicativo gratuito de inglês e português.
- [Um modelo inteligente não ajuda se não tiver para quem ligar](https://denisostapenko.com/pt/blog/a-smart-model-cannot-help-if-it-has-no-one-to-call): 2026-07-10. Cinco contêineres se dividiram entre Baltimore e Houston. O problema mostra onde termina a capacidade do LLM e falta confiabilidade ao sistema.
- [O chatbot que você usa está apoiado num artigo de tradução de 2017](https://denisostapenko.com/pt/blog/the-2017-paper-behind-the-chatbot): 2026-06-26. A indústria inteira se apoia numa cadeia curta de artigos. De Shannon ao word2vec, ao transformer, ao RLHF. Percorra uma vez e o milagre some.
- [Um grande modelo de linguagem não é inteligência artificial](https://denisostapenko.com/pt/blog/an-llm-is-not-ai): 2026-06-26. Não tem inteligência nenhuma ali dentro, é previsão de texto por uma fórmula matemática. O que um LLM é de verdade, e por que chamar de IA faz você usar errado.
- [O que realmente acontece com seu contêiner no Porto de Baltimore](https://denisostapenko.com/pt/blog/port-of-baltimore-container): 2026-01-09. 'Seu contêiner chegou' não significa que você pode retirá-lo. A linha do tempo real no Porto de Baltimore, ISF, alfândega, free time, demurrage, passo a passo.

## Português brasileiro: Pessoal

- [Olá, sou o Denis](https://denisostapenko.com/pt/blog/hello): 2026-06-22. Um breve olá, quarenta anos, construtor de sistemas, alguém que vê a IA como uma ferramenta magnífica, não um milagre nem uma ameaça.
- [Fazer quarenta anos, e recomeçar](https://denisostapenko.com/pt/blog/turning-forty): 2025-12-03. Aos quarenta parei de viver no piloto automático. Quinze anos de operações atrás de mim; agora aprendendo a construir, automação e IA. Com medo, em movimento.

## Full text and discovery

- [Full published text](https://denisostapenko.com/llms-full.txt): Both profile summaries and all published articles, with language, category, date, source URL, and complete Markdown.
- [XML sitemap](https://denisostapenko.com/sitemap.xml): Canonical pages with reciprocal English and PT-BR alternates.
- [Robots policy](https://denisostapenko.com/robots.txt): Open crawling policy.
- [English RSS](https://denisostapenko.com/feed.xml): All published notes, professional and personal.
- [RSS em português](https://denisostapenko.com/pt/feed.xml): Todas as notas publicadas, profissionais e pessoais.
- [Site information](https://denisostapenko.com/humans.txt)