https://www.loom.com/share/539543a26bbf4c5aa17ff5a4adce2ede

911Stock

Inspiration

On March 19, 2026, the DOJ charged Super Micro Computer's co-founder with smuggling $2.5 billion of AI servers to China. The stock dropped 33% in two days. By the time CNBC covered it, it was too late.

We asked: what if an AI agent watched your specific stocks, detected signals from real news sources, and called your phone before you lost money?

What it does

911Stock is an autonomous agent that monitors SEC filings, DOJ press releases, and financial news for your portfolio. When it detects something significant, it:

Analyzes the signal — matches against historical patterns in its Ghost database
Calls your phone — explains what happened, cites its sources (DOJ, Reuters, SEC Form 4), and tells you the impact on YOUR specific position
Executes trades with permission — sends an Auth0 CIBA approval to your phone. You tap approve, the agent reduces your position in Ghost DB
Learns and logs everything — every signal, trade, and decision is stored and auditable

The agent cites its sources before you ask. It knows your exact position and dollar exposure. It handles "hold on, let me check" gracefully. It never acts without your approval.

How we built it

Next.js (App Router) + TypeScript for the full-stack app
Bland AI for outbound phone calls with custom voice — the agent calls you and has a real conversation
Auth0 CIBA for trade approval — backchannel push notification, human-in-the-loop
Ghost DB (ghost.build by TigerData) — the agent's own Postgres database for watchlists, signals, historical patterns, trades, and learnings
Overmind for agent supervision and policy compliance
Gemini for signal analysis and plain-English explanation generation
Real news data from DOJ, Reuters, Bloomberg, SEC EDGAR

Challenges we faced

Tailwind v4 + Turbopack resolution conflicts required custom webpack/turbopack aliases
Auth0 middleware v4 expects specific file locations (proxy.ts + middleware.ts at project root)
Timeline playback had a stale closure bug that skipped March 19 — the most important day
Building trust: judges said "it sounds like a scam" — we added source citations, Auth0 approval, and a verified sources panel

What we learned

Trust is the #1 barrier for AI financial agents. Users want to know: how do you know? Can I verify? Who approved this? Every feature we added after user interviews was about trust — source citations, human-in-the-loop approval, audit trails.