I build
full-stackproducts &AI tooling.

75,000guests welcomed

A serverless Next.js app that takes RSVPs, sends confirmations by email and WhatsApp (through SendGrid and Respond.io), and gives every guest a QR ticket to scan at the door.

I defined the infrastructure in Terraform and deployed it with GitHub Actions, so scaling up for a crowd that size was a config change.

• Next.js
• Terraform
• Respond.io
• SendGrid

63%faster page loads

The platform serves 20,000 employees, and some of its pages were slow. I reworked the heaviest modules, the query serialization and the bulk processing, and page loads got 63% faster.

I also added transparent AES-256 encryption for personal data and built PostgreSQL workflows for the analytics team.

• Svelte
• Django
• PostgreSQL

10business units

Every unit reports its worksite incidents here and follows each one through to a fix. Managers, safety officers and investigators each need something different from it, so every role gets its own workflow.

I owned it from planning to the GCP rollout. I also set up the team's first Playwright tests and added an MCP server, so an AI assistant can work with the module through its API.

• Next.js
• Django
• GCP
• MCP

24feeds boiled down to one email

A daily newsletter for teams who want to keep up with the tools they use. An LLM reads the news, picks the stories about the tools each team uses and puts them into one short email. Each team is set up in config, and it runs on GCP, scaling to zero between editions.

• Python
• GCP
• Terraform
• LLMs

Secondsto hook up a new API

Point it at an API's OpenAPI schema and it writes the tools an AI assistant needs to use that API. A Next.js chat interface sits on top, and if one model stalls, it moves on to the next. Hooking up a new API takes seconds.

• Python
• MCP
• FastAPI
• Next.js

Soldon BOOTH, Japan's creator market

A desktop editor that finds the parts of an artwork to censor on its own, using YOLO and SAM 2.1. It can work through a whole folder in one batch and runs in English or Japanese. It installs like any other app, so nobody has to open a terminal.

• Python
• YOLO
• SAM 2.1

Publishedin a peer-reviewed journal

A university team project that predicts thermal images of pavement bricks from material and weather variables. Our work was published in the peer-reviewed JESTEC journal in 2023.

• Python
• Deep Learning