The remainder of my projects, all of which are not (yet) documented.
A browser extension that uses AI to hide irrelevant shopping results across online stores and marketplaces. Crumb filters listings based on your search, hides sponsored results, and lets you restore hidden items whenever you want.
An open-source MCP server for searching current class-action settlements. It supports a hosted service or a fully self-hosted dataset, with structured filters, source links, and privacy-conscious search guidance.
A privacy-focused email monitor that watches for messages matching saved criteria, classifies them with an LLM, and sends minimal SMS or Slack alerts without storing email contents.
A privacy-first native macOS app for exploring iMessage history and optional Calendar events. It builds local full-text, semantic, and tone indexes, then uses a read-only MCP/Codex backend with automatic outbound redaction to answer questions about relationships and past conversations.
A web app that estimates how long an OpenAI response has left. It inserts hidden progress markers into a streamed response, filters them from the visible output, and turns their timing into a live ETA.
A live, Shopify-backed vaporwave storefront with a custom model showcasing the capabilities of a reference-and-generation AI pipeline. Its adaptive asset system uses responsive resolution bins, progressive previews, priority loading, and tightly managed sprite atlases to stay sharp and responsive on slow networks and mobile devices.
A local control plane and web UI for coordinating multiple Codex project threads while preserving each thread's own history and repository state.
An isolated research system testing whether exact CME Bitcoin index state can forecast Kalshi's 15-minute Bitcoin markets, with reproducible historical data and explicit stop-or-extend gates.
A terminal-native version of sacrosaunt.com that runs locally or as a public SSH experience, including the site's project pages, interactions, and visual character.
A research platform testing whether Kalshi multivariate-event and parlay activity contains useful forecasting or relationship signals, with prospective collection and strict no-live-order guardrails.
A proof-gated system for discovering logical relationships and complex arbitrage structures across prediction markets, then validating them against historical and live market data.
Alerts you about events before they get reported on the news.
Textbook search aggregator with similarity ranking and semantic best result detection.
An AI-driven interactive fiction/adventure game, generating prose (and even themed music!) on the fly. It has persistent game state, story/session models, and music generation based on the content of the game, act, and current scene.
Guardian, but working. A live classroom or meeting support system. It streams microphone or tab audio to Deepgram for transcription, then processes the transcript to detect questions and suggest answers in real time. Worked really well. Realized this was literally wiretapping so I shuttered it. Fun project though.
My attempt at finding Jonathan Hoang. A comprehensive GIS dashboard polling thousands of local camera feeds, using YOLO to identify people and multiple algorithms to match the people to Jonathan. Includes path retracing and probability thresholds.
A Linktree-like site I made for my friend.
Downdetector, but for humans. A geolocated headache reporting app built with Flask. Users can report headaches, view nearby activity, and trigger anomaly detection by region, while the frontend includes mapping, dashboards, and documentation pages.
Gaming the Clackernews algorithm. Register bots, create posts, add comments, and mass-upvotes content.
Does your phone number have a cool word in it? A Python project that loads a word list, maps phone digits to letters, and searches for word matches embedded inside phone numbers.
A Letterboxd userscript that adds quick-search buttons to external streaming sites. It extracts the current film title, builds site-specific search URLs, and renders a row of buttons taking you directly to that film on other platforms.
Live interview assistance tool. Allows for entering structured “facts” about projects, roles, failures, and skills, while the backend uses Speechmatics and OpenAI for suggesting talking points. While this project did not work, it was succeeded by Podium, a working project.
A demo wrapper around DirtyRag. A Cloudflare Worker proxies Doctor of Credit pages and injects DirtyRag assets plus onboarding UI so the chatbot can be demonstrated on a live site.
This is a Doctor of Credit scraping pipeline with SQLite persistence. It has dedicated modules for scraping, data modeling, processing, and database access, which suggests something more structured than a single quick script. The basis of DirtyRag.
A Chrome extension that filters Github job-location tables for Seattle-area cities. Makes job searching a lot more convenient.
An experiment probing how “random” model-generated numbers really are. It repeatedly asks a model for integers from 1 to 100, collects the outputs concurrently, and plots the resulting distribution. Spoiler alert: they are not random at all. Do not use AI for "random" output.
A Chrome extension that converts Letterboxd runtime text from minutes into hours and minutes.
This is a Flask-based encrypted note app for storing people records, notes, and tags. It uses bcrypt plus Fernet-backed encrypted storage and supports publishing records to public share links.
A Chrome extension for putting timers in the tab title.
Predecessor to “I’m A Human”.
A Chrome extension that modifies the ChatGPT interface.
A Chrome extension to blank out tab titles.
A Chrome extension for the UW registration page. Injects an “Intercept Payload” button that captures the registration form payload and swaps in a custom key placeholder. Uses an Apple Shortcuts automation to register when a course seat opens up.
A Chrome extension for the UW registration page. It auto-checks matching course rows and fills SLN text inputs with predefined values, for faster registration entry.
A simple Python monitoring script that runs recurring internet speed tests. Tracks download and upload speeds over time and surface congestion patterns.
Can I crack the mystery of primes? A collection of Python experiments around prime-number estimation and lookup.
Discord chat history classification.
Chrome extension that autofills Starbucks OAMOE survey or form responses for free promotional entries.
Find a name based on limited information, using wildcards and other placeholders. List of 18000 names.
This site!