Design Engineer, AI products
I put data to work. I build the experience people use, the AI behind it, and the managed knowledge both draw on, so people, and now agents, get real answers instead of digging for them. I started on the experience side and it's still what I care about most, but I cross every boundary to get there: design, frontend, backend, data, and AI. I want it to be structurally sound and good to use.
Pattern library
-
AI platform
-
Agent UI
-
Agent knowledge
-
Design system
-
Frontend platform
-
Data viz
-
Tooling
My Work
Résumé-
JupiterOne Project Tau Rebuilt how JupiterOne builds its AI. There were about a dozen AI features, and each one shipped with its own route, agent, tools, UI, and trace. Wrote the RFC that treats AI quality as a context-building problem and judges it on four questions: correctness, relevancy, intent, and learning. Then built it: one agent loop that can stop and ask the user instead of guessing, and one tool registry shaped so MCP servers can share it. In two weeks, 13 feature-specific AI endpoints and a standalone service folded into one agent door. The agent uses the UI as its rendering engine. The app brokers every artifact the agent creates, previews, and mutates, so the user decides what lands. Tool calls are grounded in observed facts, which feed a knowledge layer with decay. That layer combines graph, account, and user knowledge, so what the system knows stays fresh and each answer builds on the last one. The aim is to move the product from pull, where users explore the graph to dig out answers, to on-demand push, where it tells them what matters and how it connects. Assistant 0.5 Assistant 0.4
-
Historical Snapshotter Gave the product a memory so it can answer a question security leaders ask all the time: are we getting better or worse? The graph only knew the present, but three platform machines had been writing daily query result rows to S3 for up to 8 years with nothing to read them. Designed and built the pipeline: capture through the existing query machinery, transcode to Parquet with DuckDB on EKS, rollup aggregation, and a report reader that serves 41 query templates. That made fast 90-day trends and executive dashboards on historical graph data possible for the first time. It's also the quantitative lane of Project Tau's knowledge. Agents call it to say what changed, not just what is. Executive dashboard Vulnerability management 2.0
-
Data Visualization Built the new app's graph canvas in-house with React Flow and elkjs, replacing a commercial graph visualization library licensed at about $69k a year. Every behavior was designed so the user can get back to where they were: packing by asset type for spatial memory, node dimming to keep context, expansions kept in the URL, and a push-pop inspector stack so exploring never strands the user. Started the chart package with D3 for the math and React for the SVG: a date brush that snaps to real data, trend charts, sparklines that break at gaps instead of inventing values, one cursor tooltip that replaced every chart's own, and a CSS-grid dashboard engine. On top of the canvas, Schema Explorer: a visual query builder for people who don't know the schema. Click asset types to build a path through the graph and watch the query write itself. The whole chain lives in the URL, so a half-built query can be shared and survives a refresh. Insights release notes
-
Experience Rebuild Made the case against our micro-frontend architecture starting in 2024: 15 independently deployed apps shared one page, so the product showed its seams and React couldn't be upgraded. When leadership approved a new app in February 2026, co-designed its layered Vite monorepo with a staff engineer, then built most of what sits on it and became its largest contributor: the design system, app shell, TanStack Router with outlet-composed routing, access gates, overrides, and preferences. The plan is one app on one shared library of forms and endpoints, with room for an AI companion next to every surface. Shipped navigation that keeps the URL as the source of truth, a universal inspector you can open from every surface so any path through the data can be shared as a link, a Cmd-K palette for every user, and a CodeMirror query editor. The first production deploy came 37 days after the first commit, and the new UI was on for everyone 148 days in. In the final push, scope quadrupled from 99 to 400 tickets and 95.5% still landed by the due date. Since the new app went live in April 2026, the frontend has caused zero incidents (10 before it), deploy failures dropped from 4.3% to 0.6%, and 82% of surveyed customers rate the new UI good or excellent. Largest contributor, with 733 of 2,123 merged PRs. Along the way, replaced a visual-testing bill that had reached $6k a month with self-hosted Storybook and Vitest interaction tests. App release notes Search release notes
-
AI Platform One of four founding engineers of the company's new AI service in October 2025. Defined the SSE streaming contract with real-time token rendering, the query validation tool, the first query-writing agent prompt, the first scored eval dataset, and an LLM-as-judge pipeline. The eval harness set the quality bar every later AI feature ships against. Replaced error events with interrupts, so the agent says what it needs instead of failing. Later added cross-region model routing on Bedrock. From September 2026, became the service's primary author, rebuilding it around one agent loop for Project Tau.
-
UX Consistency Led a push with four contract engineers to make the product feel like one product. Shipped a new virtualized data table with resizable columns and dark mode, and moved six apps onto it in a week. The rules and alerts tables went from 25 rows per page to about 500 and got faster. Engineering lead on the navigation redesign, then refactored and hardened it in the final week before launch. Started the config-driven entity inspector and shipped dark mode in the design system. Blog post Release notes
-
Bring Your Own Model Let customers run the compliance AI on their own model provider. Built the config store with sensitive fields encrypted, admin-only access with audit events, validation for each provider, and one AI client interface, so features don't care which model sits behind them.
-
Security Products Built across the security products in the legacy app. Continuous threat exposure management in 2024, as team lead from March, including an options memo that weighed five ways to compute attack paths and the risk of each. Controls and continuous control monitoring in 2025, about 60 PRs in six weeks, plus the framework wizard. Then split an overgrown repo into three products and deleted about 40k lines. In 2026, the vulnerability management backend: remediation plans grouped in three tiers, AI enrichment behind a circuit breaker, and risk scoring moved from hardcoded values to parameters. Exposure management launch Framework wizard Vulnerability management 1.0
-
Query Results and Dashboards Rebuilt query results on a new data grid behind a flag, rolled it out to customers, and deleted v1. Then rebuilt dashboard variables end to end, a first full-stack feature, in about three and a half weeks: a new service entity and GraphQL API, a query rewrite that parses the query to an AST and prunes any filter whose variable is unset (replacing string injection), and the drawer and toolbar to set them. The rewrite is selected by version, so no customer data had to migrate. Drill-through with variable values followed in 2025. New results table Matrix and drill-down
-
Frontend Guild Led the frontend guild. Set two goals per meeting, drove unique production errors from 51 to 14, set a Core Web Vitals baseline, and made "reproduce it under CPU and network throttling" the rule for latency bugs instead of "cannot reproduce." Before leading it, wrote about 30 guild standards in 2023, a dozen of them accepted best practices: container and presentational split, routable modals, URL as state, links for navigation. Then turned the rules into lint and shared tooling. Also served as incident commander for an AI service outage: caught by end-to-end tests, fixed in about four hours, with three preventive changes.
-
Design System Primary maintainer of the company design system from 2023, with 77 of 95 merged PRs that year. Shipped the new palette and tokens, then led the MUI 4 to 5 migration and rolled it across 18 repos, with old and new styling running side by side so no app had to switch at once. Designed the typed styling API and the component override wrappers, created the navigation package, and later added dark mode that stays in sync across tabs and apps. In 2025, wrote the case for standardizing patterns before components, so AI tools could understand the codebase and build in it.
-
SearchAnywhere Built two generations of global search, each shaped by what the last one showed about how security engineers search. Wrote the strategy memo and studied real queries (a quarter of natural-language searches started with "find"), then modeled the bar on a browser omnibox: history, saved questions, natural language to queries, and full-text fallback. Joined the AI search team the same year and owned the search side of natural-language querying: feedback, telemetry, retry, and the unable-to-answer state. The second generation, a docked bar with a resizable preview drawer, was announced GA org-wide in March 2025. Both gave way to the command palette in 2026. Search bar AI search Preview drawer
-
Asset Inventory Joined in February 2022, and a month in opened the 100-file foundation PR for the rebuild of the asset inventory, the product's main index, working with three other engineers. Built the filtering that compiles filter chips into queries and the critical-assets logic, then retired v1. Six months in, created the company's shared table library: keyboard grid navigation and ARIA from day one, Storybook with visual regression tests, and a render path that took 3,500 cells from 1,000 ms to 350. Critical assets
People I learn from
- Adam Argyle Modern CSS used with taste. Most of my favorite tricks here started with one of his demos.
- Martin Fowler Clear names for architecture ideas, and the habit of writing patterns down.
- Nielsen Norman Group Research-backed usability. The reason I check a design against how people actually behave.
- Edward Tufte Show the data, cut the chart junk. The bar every chart I build gets held to.
- Andrej Karpathy Treat what an AI learns as a wiki it keeps up, not a pile it searches. That idea runs through how I think about agent knowledge.