Inspiration
Active traders and small funds monitor headlines, filings, prices, and policy updates across many sources. Repeated stories consume attention, while important events can be missed or acted on before their sources are checked. We wanted to explore an agent workflow that connects a new event to its evidence, a relevance assessment, and a recorded response.
What it does
El Jerimo monitors a configured watchlist and turns fresh information into a cited assessment and an inspectable action. It de-duplicates repeated stories, adds relevant team context, checks important claims, adjusts importance using experimental category weights, and journals the event and response. Users can inspect the full trail in a native macOS interface. Actions include alerts and simulated Alpaca paper orders.
How we built it
We built an asyncio Python backend and a native PyQt6 interface. You.com Search scans 31 configured entities every three minutes, while Alpaca streams Benzinga headlines. For fresh stories, OpenCrawl searches team memory before You.com Research reads the live web and returns a cited verdict. You.com Answer cross-checks stories with importance of 0.65 or higher; unconfirmed stories are down-weighted. A learned category weight adjusts the final score. If a fresh story reaches the configured 0.70 threshold, El Jerimo sends an alert for regulators and companies or queues an Alpaca paper bracket order for a ticker. Paper orders use a 2% stop and 4% target, sized around a $200 stop risk. Live trading is locked. Every event, verdict, provider call, trade, and log line is stored in SQLite.
An experimental study uses You.com Search for dated historical stories and Finance Research for 30-day daily closes. It compares each story with the next session's move to update category weights. It is a study workflow, not proof of predictive accuracy. The walk-forward optimizer and scheduled collection of public market pages are still in progress.
Challenges we ran into
The main challenge was joining fresh information, research, and actions while keeping source provenance and action status visible. We also had to distinguish live provider evidence from mock or replay data, and queued paper orders from filled orders.
Accomplishments that we're proud of
We built a working native interface and backend for the watch, assess, act, and journal loop. The app includes a galaxy view, event feed, source receipts, a show-your-work trail, sponsor activity view, and broker-reported paper-order status.
What we learned
A decision trail is more useful to review than an isolated alert. The source, assessment, uncertainty, and action should stay connected. Clear evidence labels matter, especially when some actions are simulated.
Limits
El Jerimo is a prototype. Alpaca paper trading is simulated and does not demonstrate live execution or investment returns. The ML study does not establish predictive accuracy. Scheduled collection of public market pages is still in progress.
