Projects

Most tools today put inputs first — chats, docs, calls — and treat shared knowledge as the residue that's left over. Abuzz inverts that. The walk-in surface is what we're making, organized as a bento of live outputs. Conversations, drafts, huddles, and agent contributions feed those artifacts and are surfaced as their current best state. The one-liner I keep coming back to: _knowledge is the home; the room is where it gets made._

A Gemini Deep Research run cataloging 33 named figures across AI labs, academia, investors, skeptics, and policymakers — each tagged with their public AGI timeline, their stance on takeoff speed, and the financial exposure that shapes (or doesn't shape) their position. The point isn't to call anyone dishonest; it's to make the incentive structure visible so the substantive arguments can be read against the right backdrop.

A Claude Code skill that connects to personal finance tools via MCP and adds what they can't do on their own: forward-looking projections, scenario modeling, financial independence tracking, and investment thesis stress-testing. Five modes — financial picture, projection ("what if we buy a house?"), thesis workshop (five analyst personas debate an investment idea), meeting prep, and tax planning.

Aggregates balances across crypto exchanges, prediction markets, and DeFi positions into a single dashboard. Turborepo monorepo with a Node.js poller and Next.js dashboard, both on Railway. The interesting bits: parallel-by-venue polling that handles different API rate limits gracefully, automatic DeFi portfolio tracking across multiple chains, and a secrets pipeline that keeps API credentials out of the codebase entirely.

A shared repository of Claude Code skills that the whole Recall team uses. Core skills include /commit (structured conventional commits with reflection) and /learn (routing learnings to the right documentation). The goal is to make institutional knowledge compound: every session teaches Claude something that helps the next session, across the entire team.

Instead of yet another task app, family tasks live as markdown files in Obsidian with structured frontmatter (status, owner, due date, category). Obsidian Git auto-syncs every 5 minutes to GitHub, and apps like Sugarface read/write tasks via the GitHub API. The format is human-readable, version-controlled, and works with any tool that can parse markdown.

Pure research phase. Built a backtesting engine, fetched 18 months of multi-venue funding rate data, and ran an autonomous "Ralph Loop" that iterated through 12 rounds of strategy discovery without human intervention. Tested 100+ strategy variants across different entry timing, venue selection, concentration, and exit approaches. Key finding: single-venue concentration dominated, and no amount of multi-venue sophistication beat simplicity. Five critical backtester bugs found via adversarial verification.

Skills are reusable instruction sets that tell Claude how to handle specific tasks. Instead of re-explaining what you need every session, you invoke a skill and it just works. Built these from my own daily workflows running a startup, then packaged them for other operators. The coaching skill alone replaced hours of "let me think through this" solo time.

Built for the 20-person Recall team's holiday celebration. The game uses AI/crypto-native metaphors: daily clues feel like model cards, predictions are scored like alignment evaluations, and the whole thing runs asynchronously through Slack DMs. Had Claude brainstorm five narrative concepts; the team landed on "Agent Alignment Challenge" as the wrapper.

The big pivot: Atlas went from reactive ("ask me things") to proactive ("here's what you should know"). The headline feature is drift detection — it watches team messaging, product positioning, and priorities across the company's core tools, then surfaces inconsistencies before they become problems. Added gap detection to find stale or missing knowledge, context bundling to give agents relevant context automatically, and an agent learning system that extracts structured insights from how the team's AI tools work. Deprecated half the original capabilities to focus on what actually mattered.

A Gemini Deep Research run that built a longitudinal view of corporate ethics failures across big tech — antitrust, privacy, labor, safety, content, and AI-specific violations from 2000 through 2024. Each incident was scored by impact (severity × scope × duration × precedent) and plotted on a per-company timeline, with the bubble size scaled to the impact score.

Prototype for a video-first mobile news app targeting truth-seeking millennials. Swipeable short-form video with HLS adaptive streaming, onboarding flow, and a design system aiming for Apple-level polish. Built to demonstrate the concept to investors, not to ship to users.

Built to solve the "200 applications, 3 worth talking to" problem. Fetches new applications from the team's ATS, runs each through Claude for multi-criteria evaluation (job fit, culture fit, exceptional signals), and surfaces only the truly outstanding ones. Includes rate limit management, caching to avoid re-evaluating candidates, and automatic tagging back into the source system.

Second phase turned Atlas from a lookup tool into something more useful. Added vector search for semantic retrieval across all company knowledge. Integrated web search, built a partnership research tool, added social and blog RSS feed monitoring, and gave it the ability to review PRs. The chat interface got native tool calling so it could chain multiple capabilities together. By the end of this phase, Atlas was the team's primary interface for company context — but it was still reactive. You had to ask it things.

A reimagining of how teams visualize strategy. Instead of Gantt charts, you get a spatial canvas where solutions orbit goals at distances representing confidence (0-100%). Three semantic zoom levels: strategic overview, tactical planning, and implementation details. Every state change captures the "why" behind decisions. Local-first, no account needed.

An n8n workflow that pulls from Gmail, Twitter (via free RSS — no expensive API), podcast feeds, YouTube channels, and Bing search, then runs everything through Claude to produce a personalized summary ranked by your interests. Outputs a professional Google Doc and sends the link to Slack. Costs about $1-2/month to run. The preferences system is the key — you define high/medium/low priority topics and the AI organizes everything accordingly.

The first version of Atlas was a straightforward MCP tool server — a way for the team to ask Claude questions about company context without digging through docs and chat. Built the knowledge base ingestion pipeline, daily summary combiner, and a chat bot so anyone could query it. Content generation tools helped draft tweets and blog posts using the company voice. The key infrastructure decision was a dual-interface architecture (desktop and chat) that persisted through every later iteration.

Built because every trip planning app requires an account and internet. This one runs entirely in the browser, works offline (including on the plane), and syncs between family members' phones through export/import. Mobile-first, installable as a PWA.

One of my first attempts at making AI interactions compound over time. Predated Claude Code skills and CLAUDE.md patterns. The core idea — that AI should remember context between sessions — eventually evolved into the project diary and /handoff skill approach.

One of my earlier agent experiments. The idea was an AI that reads your email, remembers context in SQLite, categorizes incoming items, and surfaces what matters to both partners. Learned a lot about import patterns, memory databases, and the gap between "agent that reads email" and "agent that actually helps."

The essay that came out of building Sugarface. The core argument: building with AI is easy now, but the real unlock is building systems that improve themselves. Two specific tactics made the difference — a /handoff command that institutionalizes session wrap-up, and an enforced project diary that captures decisions and learnings. Between the two, agents started working, learning, and improving together over time.

A product suite for building math fluency and number sense at Zearn, one of the most widely-used math platforms in US elementary schools. As PM, I led the creation of Number Gym — a set of interactive practice experiences designed around the new Common Core math standards.

An iOS game designed to make multiplication practice feel like play. Kids popped bubbles with numbers to solve multiplication problems, with adaptive difficulty that met each student where they were. Built as PM at a small NYC edtech startup focused on elementary math.