Issue № 05Fintech AI

AI Trading Copilot

Describe a multi-leg options strategy in plain English — the copilot constructs it, validates it against risk rules, and executes across accounts.

At a glance: Ethereal built an assistant that lets a trader describe an options strategy in plain English, then builds it exactly, double-checks it against safety rules, and can place it across several accounts at once. A person always approves before anything is bought, and every price is calculated precisely so nothing is guessed.

05
Client
Confidential trading desk
Industry
FintechOptions trading · NL interfaces
Year
2025
Services
Agentic AINL Strategy ConstructionRisk ValidationMulti-account Execution
NL → tradeone conversational pipeline
Multi-legstrategy construction
Pre-traderisk validation on every leg
01

The Problem

Building and validating multi-leg options strategies is complex and error-prone. Traders need expert-grade construction and risk checks at conversation speed — not spreadsheet speed.

Overload

Traders know their intent, but translating it into precise multi-leg strike selections takes mental math and time.

Execution Risk

Fat-finger errors — a wrong expiry or a Call mistaken for a Put — can wipe out a month of profit in seconds.

Access

Junior traders grasp the theory but can't assemble complex strategies fast enough to trade safely.

02

The Solution

We built a natural-language interface that translates a trader's intent into fully-specified multi-leg options strategies, validates every leg against configurable risk rules before anything reaches a broker, and supports execution across multiple accounts.

How it works
L1Intent

Language Model

An LLM classifies plain-English intent into a strategy tag — never touching the math.

L2Rules

Risk Validation

Every proposed strategy is checked against risk rules before a human sees it.

L3Execution

Broker Bridge

Approved orders fan out to every connected account through broker APIs.

L4Core

Deterministic Engine

A Python engine turns each intent tag into exact, mathematically verified strikes — zero hallucinations.

Natural language to execution
Intent parseLLM classifier → strategy JSON
Strike constructionDeterministic · premium-based
Risk validationRule checks before sign-off
Human sign-offTrader confirms the trade
ExecutionParallel across accounts
AI for English, Python for math

The model reads intent; a deterministic engine owns every number, so no strike is ever invented.

Guardrails
Zero Hallucination

Strike prices come from a deterministic engine, never the language model.

Human Sign-off

The AI proposes the trade; a person clicks the button to execute.

Premium-Based

Strikes are selected by price, not abstract Delta — logic stays transparent and explainable.

Template Library

Execution is confined to a curated set of vetted strategy templates, not improvised ones.

Parallel Accounts

One confirmation fires the same strategy across every connected account simultaneously.

03

The Outcome

A natural-language-to-execution pipeline with domain-specific reasoning — strategy construction, validation, and execution in one conversational flow.

Built with

05 technologies
01PythonDeterministic strategy & math engine
02LLM interfaceNatural-language intent classification
03Broker APIsMulti-account order execution
04JSON tagsIntent-to-template contract
05Options enginePremium-based multi-leg construction
Next issue — yours?

Have an fintech ai problem like this?

Tell us what you're building — we'll tell you exactly how we'd ship it.

Keep reading

From the archive

All 15 case studies →