Eating Our Own Cooking

Grade My Investments' first production user is its builder — the platform runs the quarterly analysis behind the founder's own real family stock portfolio

THE VALUE LOOP

rendering diagram…

flowchart LR
    P[Fund holdings:<br/>DODGX, QQQM, TMFC] --> AI1[Claude web:<br/>decompose funds into<br/>a symbol list]
    AI1 --> SL[Paste into GMI<br/>symbol lists]
    SL --> ENG[Report engine grades<br/>every constituent on<br/>hard market data]
    ENG --> DASH[AI Report Dashboard:<br/>sort by grade]
    DASH --> TIP[The tip-offs:<br/>D's, F's, surprising A's]
    TIP --> AI2[Claude web: research<br/>articles + context on<br/>the flagged names]
    DASH --> ASK[Ask Claude inside GMI<br/>on the charts and data]
    AI2 --> DEC[Human investment<br/>decisions]
    ASK --> DEC
    DEC --> SHARE[Shared report links<br/>to family and friends]
    TIP -. rough edges found<br/>with real money .-> FIX[Platform fixes land<br/>before customers hit them]
    FIX --> ENG

The founder's own portfolio runs through the same pipeline every customer uses — and every rough edge found managing real money becomes the next fix, before customers ever hit it.

The Founder Is the First Customer

Grade My Investments wasn't built as a product in search of users — it grew out of a real need: managing a real family portfolio of individual stocks and ETFs without a Bloomberg terminal budget or a wealth manager's opacity. The platform earns its keep at home first. Every feature on this site is used — a full portfolio evaluation every quarter, after earnings seasons — with real money riding on the conclusions — which is a quality bar no test suite can replicate.

That has a name in software: dogfooding. It's why rough edges get found fast here — when the builder is also the most demanding user, "works on my machine" and "works for my portfolio" are the same test.

The Actual Workflow: AI at Both Ends, Hard Data in the Middle

The core holdings are funds — Dodge & Cox Stock (DODGX), QQQM, TMFC — which means the real question is always what's inside them, and is it still healthy? The evaluation loop, as actually practiced:

1. AI decomposes the funds. Ask Claude (web) for a comma-delimited list of each fund's holdings — seconds of work that turns opaque fund tickers into a concrete list of companies.
2. GMI grades the list on hard data. Paste the symbols into a GMI symbol list and run a report. Every constituent gets graded on audited fundamentals from the market-data pipeline — valuation, cash-flow durability, debt, growth, analyst sentiment, ML health scores. This is the step AI alone can't do reliably: numbers with a provenance.
3. The grades tip off the humans. Sort the dashboard by grade. The D's and F's — and surprising A's — are the tip-offs: a shortlist of names that have earned deeper attention, extracted from hundreds of holdings in minutes.
4. AI investigates the shortlist. For each flagged company, back to Claude on the web to search and synthesize current articles, filings coverage, and narrative context — the qualitative layer the numbers can't carry. And inside GMI itself, Ask Claude runs directly against the report's charts and data: "why is this company's health score falling?" answered from the actual numbers on screen.
5. Share the conclusion. When family or friends ask, the answer is a shared link to the live interactive dashboard — a feature that exists precisely because the founder got tired of sending screenshots.

It's an AI-evaluation approach, built as one system: AI for decomposition and interpretation, GMI's graded hard data as the trustworthy middle that keeps the AI honest. Neither half works as well alone — LLMs hallucinate numbers, and raw numbers don't explain themselves.

What Running Real Money Has Caught

Field use with consequences finds what synthetic testing can't. A reported-currency quirk in upstream market data once inflated a foreign holding's market cap absurdly — caught within hours because the number was personally unbelievable, then fixed at the data layer for every user. Share links originally opened a watered-down summary; one real send to a friend made the gap obvious, and the very next day every shared link opened the full interactive dashboard. The platform's polish is downstream of its builder's self-interest.

The honest disclosure: GMI's grades inform the founder's decisions; they don't make them. The platform is an analysis instrument, not an advisor — the same standing it has for every user, stated on every report.