Gold Desk: Agentic Analysis for a Gold Trader
Four analyst agents, a bull and bear debate and a deterministic risk gate turn live gold market data into trade ideas a trader can check, with nothing sent to a broker.
An analysis desk for a gold trader in Singapore, running on delayed COMEX prices and live central bank and market news. Macro, flows, technical and news analysts each read the data with read-only tools; a bull and a bear argue the case; a brief and, if the case is strong enough, one trade idea follows. A risk gate written as plain code sizes every idea against a limits file the trader signs. Each setup shows the odds it has had in the past, and alerts reach the trader's phone.
A trader wanted help reading gold markets around the clock: macro data, positioning, technical levels and breaking news. Language models are fluent about markets and easily overconfident, and a headline can carry instructions as easily as news.
The desk is a state machine with no agent framework. Every agent output is parsed into a contract and checked against the data before anything moves on. The risk gate is the only place a quantity is computed, and it reads limits only from a signed file. Headlines are treated as untrusted data. The desk is analysis only by design: there is no order path to a broker.
The desk runs on live delayed data with no simulated figures. A walk-forward test over two years of prices showed its odds are well calibrated out of sample, close to the actual hit rate across 3,989 setups, and also that no pattern clears costs with confidence. The desk says so on every card rather than claim an edge.
- Status
- Working system
- Client
- Confidential
Built to the Governed Cognitive Layer: the model reads and drafts, deterministic checks decide what it may do, and a named person approves anything that commits the business.
When an AI system advises on money, keep the numbers that matter out of the model. Size and limits come from code and from a file the owner signs; the agents argue and explain. Then test the advice on data it has not seen, and tell the user when the edge is not there.
Concept paper, C-suite deck and live desk
The concept paper, the C-suite deck and the working desk sit behind a password. I walk traders and risk owners through it live.
●Live demo (on request) →The same design rules behind this build, applied to your business in half a day.
Book the WorkshopDiscuss a build like thisScreens from the system
Demo data throughout. Customer names, accounts and payees are fictional.
Other projects
Agentic AIQuoteline: Conversational Sales for Small Businesses
Buyers ask on WhatsApp, Telegram or email and get an accurate, priced quotation in minutes, while a person takes over whenever money, complaints or fraud are involved.
See how it works →
Agentic AIProject Forecourt PWP: Agentic Offers at the Fuel Pump
While a driver fuels, the pump screen offers one store item at a purchase-with-purchase price, and the fuel transaction is left exactly as it is.
See how it works →
Enterprise AIKELIX Enterprise AI Platform
Proprietary private cloud AI engine built for 6,000+ engineers on multi-billion dollar projects.
Read case study →
I was asked to send alerts only for trades with better than 90% confidence. Before building it, I ran two years of replays. No setup reached 90% at an exit worth taking: to win nine times in ten, the target has to sit so close that the costs eat the profit. I said so, and we agreed on an alert for setups whose odds beat break-even instead. Telling a client the number they asked for does not exist was the most useful thing the desk produced that week.
