Mentor: a trading analysis desk run by a council of AI agents
A desk that scans the crypto market around the clock, has every setup judged by a council of AI agents (a devil's advocate, a three-vote chief) and measures every decision honestly in paper trading, never risking real money.
- 105
- decisions measured
- 3
- votes per ruling
- 100+
- automated tests
The challenge
A language model can sound convincing and still be wrong. The real question is not whether the AI has an opinion but whether its decisions beat doing nothing once costs are paid, and that has to be answered with numbers.
The solution
An analysis engine (market structure, zones, waves, multi-timeframe) proposes, a council of AI agents decides, and every veto is tracked as a shadow trade to measure what it avoided. Fees, funding and slippage are simulated, a ledger tracks the cost of every AI call, and new models only earn authority on unseen data.
The results
Across 105 judged setups, those the council approved averaged +0.30R and those it vetoed −0.16R. An encouraging result that is not yet statistically established, and reported as such.
Have a similar need?
I build this kind of solution end to end: scoping, development, production. Let's talk about your project.
Let's talk about your project