Things I'm Building

An AI-driven research partner for stock-market decision-making. It scans a ~915-ticker universe for hidden opportunities and emerging trends, fusing hard signals with self-sourced news to challenge my reasoning — not just inform it. Two modes: a proactive weekly brief that surfaces blind spots and non-obvious second-order plays, and a conversational analyst (terminal + web) that pressure-tests theses, re-runs the screen live, and plans real tax-lot-aware liquidations. Conversations persist across sessions and carry approved memory forward, so it builds on what I've already told it. Every new signal earns its place — pre-registered backtests and an evidence register keep the analyst honest. Grounded in Wake, my portfolio analysis tool, so every recommendation is rooted in what I actually hold.

A personal portfolio history and decision-analysis tool. Imports brokerage transactions across multiple accounts, computes multi-dimensional performance across a nine-tab dashboard, and replays past decisions ("hypotheticals") to turn hindsight into sharper future ones. A natural-language chat tab answers questions and draws charts on the fly. A tax-lot ledger exposes open FIFO lots per symbol and account — the data Vantage reads to plan real future decisions.

Next.js + Vercel rebuild of my personal site. JSON-driven content layer, no CMS.

A CLI tool that identifies every track in a DJ set from a YouTube/SoundCloud URL or local audio file, and prints a timestamped tracklist. It slices the audio, adaptively probes fingerprinting services (Shazam's unofficial endpoint for free, ACRCloud as a paid fallback), then reconciles the results with parsed comment-tracklists into a coverage- and confidence-annotated report. It won't beat hand-Shazaming a set track by track — but it covers hours of audio unattended, so you get a defensible starting point to fill in by ear.

Locally-run web app that fetches upcoming tour dates (for up to 12 months) for my favorite DJs. Events are displayed on an interactive map or sortable list, with time-range and proximity filters.

A semantic knowledge graph for professional network management. Uses embeddings and LLM's to support natural-language queries like "Do I know any product engineers in Europe?" and surface potential introductions.