I Build the Software
I Invest With

Flags when a company changes what it is — a reversal of stated strategy, an acquisition that changes the business mix, or a divestiture of roughly 15% or more of revenue — across a ~1,350-name global coverage universe spanning every healthcare subsector, from biotech to hospitals to tools.

In a universe that size, the most consequential announcements are often the quietest: a CEO walking back a stated strategy on a call, or a deal that quietly changes what the company is. No price screen catches them, and no one can read every call.

Uses Claude to read earnings-call transcripts and SEC 8-K filings for the three kinds of pivot, size each one, and quote the company. Every quote is checked character-for-character against the exact text the model was shown, and a finding without its evidence is rejected rather than stored. It takes no directional view — a pivot can be a long or a short.

Turns “the company changed what it is” from something noticed by luck into a systematic input — at about two cents per document read.

Answers the question every investor asks of a management team — do they do what they say? — by extracting every commitment made on an earnings call and judging it against the company’s own later words.

Management credibility is one of the most important inputs in fundamental investing and one of the least measured. It usually lives in an analyst’s memory of a few memorable misses.

Uses Claude to read years of earnings calls, extract each promise that can be checked (guidance, targets, timelines), and rule on it using management’s own later statements. Every verdict links to its verbatim quote. As of September 2026: 33 companies graded A–F on 1,159 scored commitments.

Replaces an impression of management with a record — delivered, beat, partially delivered, or missed — that can be compared across companies and over time.

Runs earnings season end to end: it keeps the calendar and consensus setup for ~1,350 companies, then delivers an analyst-grade, primary-source review before the next market open for every name I own or am researching.

Earnings season compresses a quarter of information into a few weeks. Tracking dates, pulling consensus, reading the release, the 10-Q and the call, and judging whether guidance cleared the Street — for dozens of names at once — is where analysts fall behind, and where the market moves fastest.

A daily agent keeps a calendar of every confirmed report date, synced to Google Calendar and my task list. After each print it grades the result: EPS and revenue beat or miss, whether the new full-year guide clears pre-print Street consensus, and the stock’s reaction. By the next morning’s open, a review card has read the primary filings itself — guidance versus prior-year actuals, the GAAP-to-adjusted bridge, KPIs from the release and 10-Q, buybacks versus market cap, quoted takeaways from a 19,000-call transcript library, and a verdict against my written thesis.

The first read of every print is done before the market opens — so earnings-season hours go to judgment, not data gathering.

Maintains a live ~1,350-name coverage universe that automatically detects new companies worth paying attention to — before they show up on anyone’s screen — so the analyst can decide how each one should be tracked.

A coverage list decays in two directions: existing names go stale, and new ones — IPOs, spin-offs, fresh listings — arrive without anyone noticing. Spin-offs are the blind spot: with no offering, they never appear on an IPO calendar.

Built a weekly pipeline that watches IPO registrations (including confidential filings before a listing), Form 10 spin-off registrations, and new exchange listings. Each new name is surfaced on Slack, where the analyst decides whether and how to track it, and a declined name is never re-proposed. Every name is then enriched from multiple fundamentals providers with automatic failover.

Automates the opportunity set itself. New companies arrive instead of having to be found, and the universe’s data is pushed downstream into every other monitoring workflow and scheduling tool — price alerts, earnings and event calendars, filing watches — so they all work from the same verified list.

A healthcare monitoring system that tracks policy proposals and the regularly published healthcare data on key subsectors, alongside the government rate-setting calendar that drives reimbursement.

Healthcare investing is uniquely exposed to policy and reimbursement decisions made in Washington — market-moving information that is real but almost impossible to monitor manually across subsectors.

Built three connected feeds on a healthcare-subsector spine: a running policy-news digest, an operating-conditions dashboard built from 39 industry data series, and a decade-plus calendar and history of CMS reimbursement rate cycles — all auto-published to interactive web pages.

Turns the most distinctive risk in healthcare — policy and reimbursement — into a variable that can be monitored like any other, rather than a source of surprises.

Measures whether trading actually added value — by comparing the book with what it would have done if left alone.

It is easy to remember the good trades and forget the costly ones. Without a counterfactual, there is no honest answer to whether activity helped or hurt.

At each decision date, freezes the portfolio as it stood and holds it unchanged to today, compared on a time-weighted basis against what the book actually did. Every stock sold is tracked from the sale-day close, against the S&P 500 as the benchmark — so each sell decision has its own scorecard.

Gives an objective scorecard for buy and sell decisions, separating skill from activity.

A daily drill for investment judgment: an old stock pitch from a high-quality source arrives automatically, I commit to a view on it blind — before knowing the outcome — then have a frontier LLM review what actually happened to the stock afterwards — and why.

Judgment improves through feedback, but investment feedback is slow and noisy: a thesis can take years to play out, and by then it is hard to remember what you believed and why. Without a written a priori call, hindsight quietly rewrites it.

Each morning a pitch is automatically sourced from high-signal venues — stock-pitch competitions judged by professional investors, and respected investor accounts and research sites — then delivered to a private Slack channel and printed. Before seeing any outcome, I write a structured read: what the market is missing, the key assumption and the strongest evidence for it, the valuation and return drivers, why now and which catalysts matter, the fact that would disprove the thesis, and pursue or pass with a confidence score and the next research step. A frontier LLM then researches what subsequently happened to the company and the stock, and explains which drivers actually decided the outcome — and which questions mattered most.

The point is not grading the call; it is training the instinct to find the few drivers and questions that decide an investment. It compresses years of feedback into a daily repetition — the same discipline as reviewing game film, applied to stock picking.