Engineering experiences from complex systems.Engineeringexperiencesfromcomplexsystems.

Problem

Public crime data is published as large, hard-to-navigate datasets. The challenge is turning it into something a person can explore spatially — by place, by category, by pattern — without waiting on heavy tooling.

My contribution

Designed and built the full application: the interactive map experience, data fetching and filtering flows, and the connected sidebar interactions that keep map and list in sync.

Key decisions

• Leaflet with clustered markers so dense urban areas stay readable instead of collapsing into noise
• Crime-category filters shared between map and sidebar, so one selection drives both views
• Client-side interaction model tuned for exploration: pan, zoom, filter, inspect — no page reloads

Problem

Trading platforms don't natively surface risk limits in the place traders actually work. The extension had to watch a single-page application in real time and surface profit-and-loss state without disrupting the platform.

My contribution

Built the extension: dynamic DOM monitoring for SPA environments, asynchronous P&L detection, the popup interface, and settings persistence.

Key decisions

• DOM observation designed for single-page apps — the platform re-renders without page loads, so the extension watches and re-attaches rather than assuming static markup
• Asynchronous detection pipeline so P&L reads never block the host page
• `chrome.storage` for user settings, surviving sessions and reinstalls

Problem

A professional product needed contextual AI analysis: read the elements of the page a user is looking at, generate relevant insights, and do it inside a product with real usage limits and persistence — not a demo prompt box.

My contribution

Developed the full feature across the stack: the NestJS backend integrating Google Vertex AI, the frontend feature surface built on an internal design system, persistence, and usage controls.

Key decisions

• Context built from actual page DOM elements, so model input reflects what the user sees
• Vertex AI integrated behind a service layer with explicit limits and controls, not raw calls from the client
• PostgreSQL persistence for insights and usage accounting; frontend shipped inside the product's existing design system