What whales, insiders, and Congress are actually buying — in one terminal.
Retail investors now account for roughly a fifth of U.S. equity volume, and most of that growth is recent — tens of millions of new brokerage accounts opened in the last few years alone. The trader AlphaGraph is built for isn't reading 1,000-line-item sell-side models the way an equity analyst does. They're watching for unusual options activity, because large institutional bets often show up in options volume before they show up anywhere else, and following FinTwit, Reddit, and podcasts for early signal. It's also, per aggregated brokerage data, a trader who has collectively lost billions of dollars on options — largely by buying volatility instead of understanding how to manage it, and getting caught by IV crush around earnings.
None of the underlying data is actually private. Form 13F filings, Form 4 insider trades, STOCK Act disclosures, options flow — all public, all regulated. What's expensive is real-time infrastructure: live options data alone carries a $1,500/month exchange redistribution fee plus a separate $2,000/month charge just for calculating anything with it on a backend, before per-user fees. So the build was scoped deliberately around that economic reality — a first version built entirely on historical and end-of-day data, which carries none of those licensing costs, sized for the swing and position traders who make up most of retail activity rather than the minority trading intraday.
AlphaGraph aggregates Form 13F whale portfolios, Form 4 insider buying, Congressional STOCK Act disclosures, unusual options flow, and now full supply-chain relationships into one workspace — and tailors what it surfaces, and how it explains it, to the person looking at it.
The gap wasn't access to the data. It was that nobody built the version meant for one person, not a desk.
The starting point was a persona, not a feature list: who actually uses a product like this, and what does their process already look like?
That trader mostly isn't reading fundamentals the way an analyst does — they can't realistically out-model Wall Street on a discounted cash flow. What they actually do is scan for unusual options activity as an early signal of institutional conviction, and follow social platforms for the ideas that don't show up in a screener yet. Options are also where they get hurt: retail traders overwhelmingly buy volatility instead of understanding how to manage it, which is a structural reason so much retail options activity ends in a loss around earnings specifically.
The second half of the research was economic. Real-time options data (OPRA) carries a $1,500/month redistribution fee plus a $2,000/month non-display fee for any backend calculation, before per-user charges — and direct exchange connectivity runs five figures to set up and five figures a month to maintain. Historical and end-of-day data carries none of that. So the roadmap was staged on purpose: a Phase 1 built entirely on T+1 and EOD data, under a hard sub-$100/month data budget, sized for the swing and position traders who make up most of retail — with real-time flow and dynamic options Greeks scoped as a deliberate Phase 2, once there's a reason to carry that cost.
I directed an AI coding agent through the build rather than writing every line myself — which meant the real work was deciding what belonged in the product before a prompt was ever written.
Chose the data sources that actually change a self-directed trader's decisions — whale positioning, insider activity, congressional trades, options sweeps, sector rotation, supply-chain relationships, nightly synthesis — instead of trying to cover every institutional feed that exists.
Trading horizon, risk tolerance, sector focus, and active watchlist as the variables every AI-generated briefing and alert gets filtered through, not an afterthought bolted on top.
Worked through the agent across the dashboard, module views, and briefing-generation logic, reviewing every UI and data-shape decision it proposed against the product spec.
Iterated the nightly-briefing prompt structure until it produced specific, persona-calibrated trading notes instead of a generic market wrap anyone could read.
The shape is simple on purpose: pull from primary regulatory and corporate sources, reconcile and store them, then let AI synthesize — never invent — what's already verified.
This is the Phase 1 architecture — built entirely on historical and end-of-day data to stay under a sub-$100/month data budget. Real-time OPRA streaming and live Greeks are a deliberate Phase 2, staged for once the cost is actually justified, not a missing feature.
Built with
The personalization engine is the part of the product that's easy to skip and hardest to get right. Every setting here — swing trader vs. long-term investor, conservative vs. high-alpha, which sectors matter — is a direct input into what the AI decides is worth telling that user that night.
The other constraint was literacy, not just personalization. This trader doesn't necessarily know what “gamma exposure” or “risk-on regime” means the first time they see it, so every module ships with a plain-English guide built directly into the interface: what the feature actually is, why it matters to a retail trader specifically, and a short numbered walkthrough of how to use it — not a glossary buried in a help center.
Holdings data answers what a company owns. It doesn't answer a question that matters just as much to a real thesis: what does this company depend on, and how exposed is that dependency? The Ecosystem Graph decomposes a company's actual supply chain and partnerships — tier by tier, sourced from 10-K filings and press releases — into something a retail investor can read, instead of something buried in a footnote three levels down.
Every relationship gets an asymmetry & moat score and a bull/bear read specific to that relationship, not a generic take on the ticker. And every node carries its own verification tag, because a supply-chain map is only useful if you can tell which edges are confirmed disclosure and which would just be the model guessing.
The nightly briefing is the part of AlphaGraph most exposed to hallucination risk — it's the one module where an LLM is actually writing analysis instead of just displaying structured data. A model asked to write a market briefing with no constraints will confidently invent a yield move, a catalyst, a rotation that didn't happen, in the same fluent tone as one that did.
So the briefing generator isn't allowed to free-write. It's fed the actual reconciled data for that session — real treasury yields, real sector performance, real 13F and options-flow numbers pulled the same day — as its only source of truth, and constrained to a fixed output structure: regime tag, core catalyst, sector winners and losers, execution rules. The Ecosystem Graph runs on the same discipline: every moat score and bull/bear read is tagged against a specific confirmed disclosure, not generated freehand. The model's job everywhere in the product is to synthesize and prioritize what's already verified, not to generate new claims.
AlphaGraph is live, spanning three onboarding tiers — Simple Prep, Catalyst Flow, and Full Terminal — across whale 13F consensus, Congressional STOCK Act activity, options-sweep flow, sector rotation, and now supply-chain decomposition, all in one personalized terminal.
Beyond the product itself, it's proof that the research holds up: a version of this intelligence built entirely on public, historical, and end-of-day data can still be genuinely useful to the retail trader it was designed for. It's a working example of a discipline that runs through every module: understand who you're building for and what data reality you're working with before writing a line of the product, direct an AI coding agent to build it, and stay accountable for whether what ships actually helps.