90%↓
sprint planning time
Visual context for agents
Senior Software Engineer · Fullstack & AI Systems
I build the platform, not just the feature.
Six years shipping financial products — from embedded POS terminals at Stone to banking PWAs used by thousands of JP Morgan Chase clients. The pattern repeats: I find the problem underneath the problem, and build the thing that solves it for everyone.
Measured outcomes
80%↓
end-of-sprint bugs
Agent-assisted TDD + regression
60%↓
token spend per task
Context engineering
70%↓
sprint time at JPMC
In-house component library
Selected work
Things nobody asked for. Things everybody used.
None of these were assigned. Each started as a problem nobody had named yet.
AI Platform 2025
MCP Server for design tokens & components
Venturus
Agents kept inventing components that didn't exist. So I gave them the real thing: an MCP Server exposing design tokens, patterns, components and already-rendered elements as tools — the project's actual truth, queryable.
Agents now build features, fix bugs and write regression tests with real project context instead of guesses. Hallucination dropped; onboarding for new features got 40% faster.
Agent Tooling 2025
Screen annotation for agent context
Venturus
Captures navigation screenshots with on-screen event markers in COCO format, feeding LLMs and SLMs visual context instead of paragraphs of description.
Cut sprint planning time by 90% and token spend by 60%. Bug reports and prototype-to-code translation went from prose to pixels.
Fullstack + Agent 2025
Legacy flow assistant
Venturus
The client had a business rule buried in a legacy system nobody could trace day to day. I built the app that explains it — and walks you through it.
Node/TypeScript, Express, PostgreSQL, Zod, SSO, Docker. Svelte front-end driven by a LangGraph agent that consumes the MCP Server and design system to render components live. Flagged by leadership as a potential product.
Developer Platform 2022–2025
API documentation engine
Ília Digital · JP Morgan Chase
Hundreds of endpoints, documented as diagrams nobody could keep current. I proposed generating docs from the APIs' own .yaml files, with Web Components demonstrating real usage and flow.
Built the MVP; it was adopted and became its own project, which I led. Replaced Docusaurus, cut first load by 60% and script execution from 11s to 1s. Core shipped as an NPM package reusable across React and Svelte.
Embedded Systems 2021–2022
Containerized apps on POS hardware
Stone Payments
Docker containers running on the payment terminal itself — not on a build server. Every megabyte of image had to be earned.
Trimmed image, dependencies and runtime footprint to fit embedded memory and storage limits. In exchange: isolation and predictable deploys across a fleet in the field. Small-screen defects fell 70%.
Track record
Six years. Mostly inside banks.
Venturus
Senior Software Engineer
R&D center for global tech companies
Jun 2025 — Present
The team's go-to engineer and the tech lead's right hand. Since generative AI landed, the team's go-to for AI systems engineering.
- Architected the MCP Server, the design system and the agent stack they feed.
- Automated code review, task management and metrics through AI workflows.
- Mentor a junior developer; support the team on requirements and technical decisions.
Ília Digital
Senior Software Engineer
Client: JP Morgan Chase
Apr 2022 — Apr 2025
Banking and investment platforms used by thousands of JPMC clients, built while embedded in the bank's teams.
- Led the API documentation engine, from unsolicited MVP to its own project.
- Python APIs (RabbitMQ, PostgreSQL) for asset management; Elixir APIs for corporate cards.
- Svelte component library replacing Material-UI — 70% less sprint time.
- PWA for expenses, transfers and property investments — 40% faster loads.
Stone Payments
Software Engineer
Embedded POS terminals
Sep 2021 — Apr 2022
Applications running on payment hardware, where every KB of bundle and every CPU cycle sits between a customer and a transaction.
- Shipped Docker containers running on the POS device itself.
- Authentication and security flows to financial-industry standards.
- Mentored interns; active in refinement and prioritization.
Decode
Full-Stack Engineer
BTG Pactual marketing startup
Apr 2020 — Sep 2021
One of the lead developers, working database-to-interface in direct contact with marketing, design and BTG Pactual clients.
- Content trends dashboard: Node/MongoDB aggregations through to D3. APIs served 10k+ daily requests.
- Internal file platform for BTG's executive board — sensitive-data context, upload and permissioning API.
- Defined the JWT auth strategy for the bank's internal apps: issuance, refresh, role-based access.
- Landing pages at Lighthouse 95+, cross-browser down to legacy Internet Explorer. Chose SSR or SSG per project.
Tools
Deep in a few. Fluent in the rest.
AI Agents LangGraph MCP Servers & Clients RAG Embeddings Context Engineering Vector DBs LangChain Fine-tuning
Next
Got a problem underneath a problem?
I'm open to conversations about AI systems, internal platforms and engineering that makes whole teams faster.