guide
Hi, I'm João. Ask me about my work by text or voice, or let me walk you through the portfolio myself.
Projects
These are the projects I want people to judge me by: what they do, what I built, the tradeoffs I made, and the stack behind them.
Aurora started as a local voice assistant and grew into a privacy-oriented capability mesh: typed service contracts, LocalBus/BullMQ process modes, WebRTC peer calls, recipient-specific permissions, shared React/TypeScript clients, Rust/WASM authorization, native speech, and Android lifecycle handling.
A deep AI systems project: voice UX, local privacy, tool orchestration, service contracts, RAG, model-provider flexibility, and distributed mesh architecture all visible in one codebase.
Discrypt explores a Discord-like secure communication system where Rust owns sensitive key/media boundaries, relays route ciphertext without trusted custody, deterministic harnesses model delivery/recovery behavior, and the interface states the limits of content privacy and deletion.
A security and distributed-systems artifact with protocol-aware frontend work, clear product boundaries and systems engineering beyond a normal UI prototype.
Nakamura Advogados is a shipped client-facing law-firm website built around page structure, visual hierarchy, responsive implementation, restrained professional tone, live-site presence and SEO-oriented frontend work.
A shipped frontend example showing client delivery, visual polish, SEO basics, and business-facing information architecture.
Secure Env Manager creates Distrobox-based development homes backed by LUKS storage, masked host-home access, reduced capabilities, isolated credentials, browser/editor separation, controlled audio/video/X11 bridges, and encrypted export/import across machines.
A concrete AI-security tooling project: each agent context gets its own encrypted storage, browser profile, editor state, SSH identity, and filesystem boundary.
Tatovering pairs a frontend with a Go API for a tattoo-scheduling prototype using Next, Ant Design, Tailwind, i18n, studio discovery and MapBox markers. The API presents Go/Gin/Supabase CRUD plus agendamento routes.
A product-engineering example covering fullstack CRUD, marketplace modeling, booking UX, and frontend/backend integration.
Incubator Manager is a local microservices platform for the ILB incubator domain: ten Django REST backend services, Next.js frontend, Nginx gateway, Postgres, Redis, RabbitMQ, MinIO and Make/Tilt workflow.
A production-shaped strict-microservices system runnable locally with one command: clear service boundaries with no shared ORM models, idempotent event consumers, gateway-injected identity headers, Tilt live development, and a full host-side quality gate spanning ten Django services and the frontend.
Brain Signal Processing Game is a neurofeedback research software project connecting OpenBCI/LSL EEG acquisition, Python processing, Welch PSD beta-feature output and a Pygame feedback loop.
A concrete brain-computer interaction artifact: EEG acquisition, signal processing, asynchronous data transport, and an interactive neurofeedback game connected into one reproducible research software loop.
I built ABCD from scratch at Adopt AI as an open-source coding-agent workspace for turning HAR/API recordings into WDL actions, tests, drafts, and publishing steps.
In an Odoo integration demo, I took an integration task from 7 days to 2-3 hours.
Execution infrastructure for trusted functions, generated code and long-running workloads.
I helped design and develop Tabby with the Adopt AI team: browser sessions for human login handoff, per-session worker pods, authenticated API execution, and browser automation.
Human login can hand control back to automation inside the same authenticated session. Prepared recording sessions move browser startup work ahead of the request, while single-use cleanup prevents a used session from returning to the pool.
A Python/LangGraph insight-generation service for executives who needed recommendations, not only direct answers to known questions.
Reframed the analytics product from query answering toward recommendation generation. Grounded model reasoning in customer-specific business meaning rather than only schema retrieval.
A Canvas/Fabric.js editor that turned physical venue layout into stable seat records for organizer editing and buyer selection.
Frontend/product work where geometry, interaction and domain identity matter together. Connects organizer layout editing directly to buyer seat selection.
I built this site so you can explore my work through diagrams, questions, narrated tours, and voice. The same Supabase content drives the pages, guide, navigation, citations, search, and tour targets.
The site can move between reading, graph exploration, guided tours, search, and speech while each path reads the same portfolio content.
OCR and specialist agents turn handwritten forms into structured spreadsheets and filing documents.
Separated perception from interpretation: OCR and segmentation handled visual extraction, while agents imposed business structure. Produced structured spreadsheet and tax-form/PDF artifacts for human/business review.
Forecasting and AI recommendations support inventory orders and transfers with human approval.
Demonstrates FDE work across customer systems, ML, LLM reasoning, and effectful business actions. Human approval is part of the architecture before orders or transfers are initiated.
Voice guide
guide
Hi, I'm João. Ask me about my work by text or voice, or let me walk you through the portfolio myself.