Why I Build Investment Tools as a Side Project
Institutional investors have Bloomberg terminals and quant teams. Retail investors have Reddit and hot takes. I built an AI research copilot to close that gap — here's why the problem is worth solving.
The Gap Nobody Talks About
Every time I open a brokerage app, I'm struck by the same feeling: the information design was built for someone who already knows exactly what they want. Here's the earnings date. Here's the chart. Here's a news headline from six hours ago.
What's missing is *synthesis*. Nobody is helping me understand whether the earnings report actually changed the thesis, whether the options market is pricing in risk I should care about, or whether the debt-to-equity trajectory over five years tells a different story than the current quarter.
Hedge funds have that. They have quant teams running factor models, analysts reading every 10-K, and proprietary data feeds from alternative sources. A solo retail investor doesn't have any of it.
That's the gap I started building around.
SEC EDGAR Is Hiding in Plain Sight
Here's something surprising: the SEC makes years of financial data freely available through its EDGAR API. Every public company files 10-Ks, 10-Qs, 8-Ks — and all of it is machine-readable if you know how to ask for it.
The problem isn't access. It's that the data is spread across thousands of filings, uses inconsistent terminology between companies, and requires serious financial domain knowledge to interpret correctly.
When I started building the AI Investor Platform, my first milestone was just answering the question: can I reliably pull five years of revenue, net income, free cash flow, and debt from any S&P 500 company using just the EDGAR API? It took about three weeks of iteration to get a reliable parser that handles the edge cases — things like companies that change their fiscal year, subsidiaries with separate filings, or financial restatements.
Once I had the data, I started calculating ratios automatically: P/E, EV/EBITDA, ROE, debt-to-equity, current ratio. Twelve metrics in total. Not because twelve is the right number, but because those are the ones that show up consistently in the research frameworks I trust.
Why LLMs Actually Help Here
The first version of the platform was just tables and charts. You'd pick a ticker, and it would show you the financial data in a clean layout. Useful, but not transformative.
The second version added language model integration. And that's where things got interesting.
The insight is that LLMs are genuinely good at pattern-matching across large documents and synthesizing that into natural language. They're bad at math and shouldn't be trusted with raw calculations — but they're excellent at reading a 10-K and identifying what management is signaling between the lines, or comparing the risk factors section year-over-year to see what's new.
So the current architecture is a hybrid: structured financial calculations done in code (no hallucinations), and language model interpretation layered on top. The model doesn't calculate the P/E ratio. It reads the cleaned financial data and tells you whether a P/E of 34 makes sense given the company's growth trajectory and competitive position.
Ten Power Users and What I Learned
The platform is in private alpha with ten users. All of them are technically sophisticated — engineers, a financial advisor, a couple of active options traders. The feedback has been humbling.
The thing people value most isn't the automation. It's the consistency. When you're doing research manually, your process varies by day — how much time you have, how much caffeine you've had. The platform forces a consistent framework every time: same metrics, same timeframe, same structure. That consistency itself is a competitive advantage.
The thing people value least is the generated summaries. LLM prose has a specific quality — confident, structured, slightly generic. Everyone ignores the summary and goes straight to the data. I'm iterating on this.
What I'm Building Next
The next phase is options integration. Right now the platform handles equities well, but options research requires a different layer: implied volatility surfaces, Greeks across strikes, historical IV vs. realized volatility comparisons, unusual flow detection.
That's what the AI Options Tracker is for — a separate tool that focuses specifically on the derivatives layer, with agent-assisted screening and trade idea generation.
The long-term goal is something closer to a unified research environment: equities fundamentals, options flow, and macro data in one place, with agent workflows that can run background research and surface relevant changes automatically.
It's a long road. But the gap is real, and nobody else is solving it for the retail-but-sophisticated investor.