TrumpTradesBuyTheDip

Project story: TrumpTradesBuytheDip

Hackathon track: Money Intelligence
Team: AlwaysLateToEverything Team Members: Frank Ivey

Why I built it

Public officials’ financial disclosures can be hard to read and arrive long after a transaction. I built TrumpTradesBuytheDip to turn a reported trade into a clear, source-linked story: who owned the asset, when the transaction happened, when the public learned about it, and what a price change actually means. A beginner should be able to see the evidence and the delay before treating a headline as a trading signal.

What the app does

The live hackathon app presents three fictional sample filings. Each card keeps the filer separate from the disclosed owner, shows the transaction and public dates, preserves the reported amount as a range, and opens the exact sample source row. A sample price watch shows its reference price, observed price, calculation, and timestamp. It describes a change in price; it does not recommend buying or claim to know the official’s profit. Users can ask a question by typing or speaking, read the answer, and hear it spoken by their browser. The app does not connect to a brokerage or place orders.

How I used Nicrron and the model

All live model calls go through Nicrron. I selected openai/gpt-4o-mini through Nicrron because this prototype needs a short, dependable classification of a beginner’s question, not a model-generated trade claim. The model chooses one of five topics: owner, timing, amount, price, or source. Python then builds the answer from validated sample fields and the filing evidence. Unexpected model output is rejected, and a factual sample summary remains available if the model call fails. The browser’s speech tools handle audio; the model does not synthesize speech.

How I used PRISM

I uploaded the fictional filing to PRISM Knowledge Base, assigned it to the app’s agent, and search it before explanations. The app sends PRISM a trace with the question, answer, model, and session, plus spans for the Knowledge Base retrieval and Nicrron call. I also replayed six deliberately flawed sample cases that confuse filer with owner or filing date with transaction date, then six corrected cases on the same questions. This gives a visible before-and-after trail.

The Knowledge Base and trace integration is verified in the live app. The other two Money Intelligence proofs are not yet complete: PRISM’s trace analysis reported an upstream authentication error, so Root Cause created no failure cluster, and the current account returned HTTP 402 when creating a custom evaluator. I will only claim those proofs after PRISM produces the hosted cluster and same-input evaluator scores.

What comes next

A real-data version would need authorized access to official disclosures, careful handling of amendments and reporting delays, and licensed price history. It would track hypothetical asset returns from both transaction and first-public dates, never infer a person’s exact realized profit from a disclosure range. Robinhood trading is outside this sample-only hackathon app and would require an approved integration.

Live demo: trumptradesbuythedip-demo.onrender.com
Source code: ithllc/TrumpTradesBuytheDip (AGPL-3.0)