goEscargot FX Hedging

Architecture Diagram

DEMO LIVE at https://demo.goEscargot.com

Inspiration

Any company that sells or buys in other currencies has to deal with exchange volatility. Our product solves this.

The idea for goEscargot came from the pain point of FX hedging practice at work. In FX hedging, every treasurer struggles with the same questions — when to hedge, for how much, and for how long. Even with a solid policy, timing and layering decisions are still guesswork, and tracking them in spreadsheets is chaos.

What it does

goEscargot automates that entire process. It uses AI to combine historical FX data, company policy, trading rules, market trends, and news to recommend the optimal timing, hedge ratio, and tenor. It connects with ERP for FX exposures, then sends trades to FX platforms for executions, and tracks trades automatically, giving treasurers a single, reliable dashboard instead of huge spreadsheets.

How we built it

We've developed a multi-tier app on Google Cloud, in particular Cloud Run and the new Agent Development Kit for delivering a powerful multi-agent AI.

ADK Multi-Agent on Cloud Run (Python)

Google ADK enabled us to rapidly develop an agent server connected to Gemini API. A key powerful feature of the framework is how easy it is to connect tools and other agents, to create a powerful AI system. We detail the agents below. The Cloud Run service runs internally, only our app server needs to connect to it.

App Service on Cloud Run (Node.js)

A Cloud Run service runs the rest of the app on Node.js. Data is stored on Cloud Buckets mounted to the instance. This service sends company data (rules, policy, status) and FX data (historical, realtime) to the AI Agent. The AI agent returns a recommendation that is rendered in the frontend Dashboard.

Data Sources: Historical data is fetched the Bank of Canada Valet API, an open api with full daily currency rates. For moment to moment rates, we fetch from the Y! API.

Data Storage: FX rate data caching, and the company rules, policy, exposure, and hedging history, are all stored in the cloud in buckets.

Notifications: We send emails to the user. Currently using Brevo. We are also integrating with Trading platforms.

Frontend: We used light Framework-less javascript, with sophisticated component and state patterns leveraging Server-Side rendering in Node.js.

Challenges we ran into

Implementing a custom Hedging Rules system was an our first dilemma. We had to design a system that allowed for rules that fit the company objectives and risks. We did this by building dynamic rule “scenario” models. Based on historical and current FX rate data, the companies customized scenario is determined and a recommendation is made based on that.

We quickly found a limitation of hedging rules: we lacked a broader market and financial context to make the best recommendations. Now more than ever, world events can impact currency markets moment to moment.

The solution: a financial advisor AI Agent! Google’s Agent Dev Kit, leveraging multiple agents, was perfect. We need multiple scopes of consideration:

Multi-Agent Architecture

The system uses 4 specialized agents coordinated by an orchestrator:

1. Market Analysis Agent

Searches recent currency (ie. USDCAD) news using Google Search
Analyzes market sentiment and volatility
Identifies key economic events (BoC/Fed meetings, GDP reports)
Provides trend analysis and risk factors

2. Strategy Agent

Generates forward contract strategy and pricing
Ranks strategies by risk/reward profile
Applies company policy constraints

3. Risk Assessment Agent

Calculates Value at Risk
Validates policy compliance
Assigns confidence levels to recommendations

4. Orchestrator Agent

Coordinates the three specialized agents
Aggregates outputs into final recommendation
Compares AI suggestions with threshold-based rules

Accomplishments that we're proud of

After just a couple weeks of coding, the recommendation system and dashboard are usable. The speed and power of Google ADK is remarkable. The product delivers real value to companies. We think there is a real potential to go forward.