nikolasdehor
Full-stack engineer turning AI into software that survives contact with reality.

Portfolio · DeHor Labs · LinkedIn · Email

I build the layer between an AI model and a real workflow: agent runtimes, MCP servers, RAG pipelines, multimodal knowledge systems, automation, and full-stack products. I like systems with a clear failure mode, observable decisions, and a way to prove what happened.

I am COO and co-founder of FOR6 Solutions and founder of DeHor Labs, based in Goiânia, Brazil.

messy input → tools + retrieval + memory → agent decision → useful software

• AI agents that can call tools, remember context, recover jobs, and hand work back with evidence.
• MCP servers that expose Brazilian fiscal, agribusiness, and legal workflows to assistants.
• Multimodal pipelines for audio, video, images, OCR, embeddings, and source-backed RAG.
• Full-stack products: commerce, healthcare operations, accounting, affiliate platforms, data products, and mobile apps.
• Engineering guardrails: visual QA, security review, CI gates, deploy risk, and operational memory.

These are the products and public surfaces that best show the range of my work:

• Forgiarini Parfum — full-stack perfume commerce: catalog, search, cart, WhatsApp checkout, Supabase-backed admin, product data, and responsive storefront.
• PlantonIA — AI WhatsApp assistant for healthcare office administration, with conversation context, scheduling, audio transcription, escalation, and tenant isolation.
• AgroVoz — multimodal rural assistant for voice, text, and photos, with semantic retrieval over farm operations (closed beta).
• Atlas AI / Atlas Contábil — accounting automation product in progress, with an agent, asynchronous extraction, WhatsApp workflows, and client portal.
• Saúde Sem Espera — Flutter healthcare queue and service-management product with mobile app, panel, and public invitation flow.
• RifaZap — full-stack raffle product with Pix, reservations, admin, cryptographic draws, and WhatsApp flows.
• LucraLink — affiliate and subscription SaaS with click tracking, commissions, Telegram automation, and payments.
• Licitações.dehor · Patrimônio · Sócios do poder — public-data products built around ingestion, provenance, semantic search, and reproducible methodology.
• SaborIA — multi-agent RAG system for intelligent restaurant-menu analysis.

• MCP Fiscal Brasil · Docs · PyPI — Brazilian fiscal MCP with CNPJ, NF-e, NFS-e, CT-e, SPED, eSocial, Simples Nacional, and tax-reform data.
• MCP Agro Brasil · PyPI — agribusiness prices, weather, exchange rates, exports, news, and crop calendars.
• MCP Jurídico Brasil — modular legal MCP over public CNJ/DataJud data.
• TranscreveAI — video and audio into searchable multimodal dossiers, frames, OCR, embeddings, citations, and RAG.
• Verboo Bridge — controlled routing and execution of AI subagents through MCP.
• Agent Bridge(private/local) — delegates work to any installed AI CLI as a subagent; kept without a public link because the repository is not public.
• Free Models Monitor — tracks free-tier LLM changes and proposes fallbacks for agent harnesses.
• Visual Eyes · Semtree · ShieldCode — visual inspection, semantic code context, and security hardening for AI-assisted development.
• OpenParty — self-hosted synchronized watch party with WebSockets, chat, and reactions.

I also work inside existing ecosystems. The wording matters: these are contributions, integrations, forks, or operations around upstream projects — not claims that I created them.

• OpenClaw and Hermes Agent: runtime integration, messaging gateways, automation, and operational work.
• nanobot, PicoClaw, VibeTunnel, AIOS, and BMAD: reviews, fixes, pull requests, and engineering feedback.
• Canship: deploy-risk gate that uses a repository's own failure history before merge.

• Cast Group / Banco do Brasil: I work on full-stack engineering, AI-assisted documentation quality, automation, generative-AI tools, and CI/CD in a corporate banking context. Internal repositories, metrics, prompts, and architecture are intentionally not published here.
• LUME: I am rebuilding an enterprise agent hub with multi-agent orchestration, RAG, embeddings, scalable memory, MCP, AWS, and full-stack services. It is an active project in development, not a public product claim.
• Rubiataba Industrial: I built operational automations, including intelligent monitoring and recovery for sugar-weighing scale services, plus APIs, bots, cloud deployment, and infrastructure support.

• Agent memory, RAG, embeddings, hybrid retrieval, and evidence-backed context.
• MCP surfaces for Brazilian business, fiscal, legal, and operational workflows.
• AI code review, visual regression, security hardening, and CI/deploy feedback loops.
• Products that connect WhatsApp, APIs, databases, queues, cloud services, and humans.

Python · TypeScript · JavaScript · Go · React · Next.js · Flutter · FastAPI · NestJS · PostgreSQL · MongoDB · SQLite · Redis · Supabase · Docker · AWS · GitHub Actions · MCP · RAG · Whisper · OCR · Playwright

Goiânia, Brazil · dehor.dev · news.dehor.com.br · LinkedIn · nikolasdehor79@gmail.com