Why I built
The Prompted Portfolio.

I didn’t have time to invest the way I wanted to

Notes, spreadsheets, and saved sources all need keeping up. When that upkeep takes the time you have, forming a view, testing it, and deciding what to own get squeezed.

For years after my last active-trading role ended in 2017, I mostly wished I had time to manage my own portfolio actively.

What AI made practical again

I recognized ChatGPT’s potential for portfolio management in 2022, but I did not resume active management then. What changed that came in pieces: conversations about world events, writing code again outside work, and automation I built to help my father-in-law. With chat tools I could argue a view through. A coding agent built the backtester I now use for my quantitative strategies.

In one recent conversation, an AI response connected my separate ideas about memory supply and power to a possible shared dependence on capital spending by large cloud and computing companies. It also suggested conditions to monitor. I used that response to narrow the next research questions.

The experience behind the method

From 2005 to 2013, I worked on portfolios for ultra-high-net-worth clients, as an analyst until 2008 and then as a portfolio manager running $2 billion in assets. I also spent two years developing automated strategies as a proprietary trader, and about ten years in software engineering and site reliability engineering. Most recently, that engineering work included a global Kubernetes platform spanning more than 50 datacenters.

I began an equity process in 1997 using fundamental screens, filings, company outlooks, and conviction-based weights. By 1999, that process had won portfolio contests and brought me paid portfolio-management work, while I was barely out of college. From 2013–2015, proprietary trading taught me how much the mechanics of a trade, such as spreads and order types, can decide whether a strategy works. From 2015–2017, I ran a customer asset-allocation system, first at a startup and then inside Goldman Sachs after it acquired the company; the system rebalanced every day to put customers’ new cash to work.

How I work now

I use AI conversations to sharpen a question and argue against my answer, then go to the filings and data to see whether the answer holds. For stocks, that means finding companies exposed to a thesis, comparing their ratios, and reading the filings of the most promising ones.

For quantitative strategies, I use a backtester that a coding agent built to follow my methodology. I enter the variants I want to compare. It runs screen backtests, rolling backtests, and simulations, then combines the strategies’ returns into a portfolio itself. For each variant it calculates a deflated Sharpe ratio, which marks the result down for how many variants I tried. Then it shows the variants side by side and saves everything in a study I can update with one click. The version I’ve used so far draws its data from other providers. As of October 7, 2026, a version that uses Portfolio123’s data is in development.

I’m also building a research application around an AI agent. I’m happy with how it reads SEC filings, and I’m extending its analytical tools, including how it uses Portfolio123’s data.

What this isn’t

This is education about using AI in your own investing, not personal investment advice. I don’t promise returns, and nothing here tells you to buy or sell anything. I plan to publish my own portfolio’s decisions once that record is running.

How I keep it honest

Every guide says whether its example is real or invented, where AI helped, and what I’m still unsure of. When I get something wrong, I fix it and say so on the page. AI helps with research and drafting. I pay for my own ChatGPT and Claude subscriptions and have no affiliate or business relationship with OpenAI or Anthropic. I review everything and I’m responsible for what’s published.