FX rate intelligence for the Treasury team
FX rates arrived in dozens of WhatsApp and Telegram groups, so Treasury had no single view of who was offering what. I built an AI system that turns those messages into structured data, finds the best rates and flags arbitrage.
- Saved every day for Treasury
- 3+ hoursSaved every day for Treasury
- Provider and partner sources monitored
- Up to 50Provider and partner sources monitored
- Finance and Treasury people on the Slack reports
- 3 usersFinance and Treasury people on the Slack reports
The problem
Treasury decisions depend on knowing the best rate on offer right now. Those rates were scattered across many group chats, in free-form messages, so spotting the best provider or an arbitrage window meant watching dozens of chats by hand.
What it does
- Collects unstructured FX-rate messages from many WhatsApp and Telegram groups.
- Uses AI to turn each message into analysis-ready rows in Google Sheets.
- Sends reports to Slack for the Finance and Treasury team.
- Answers on-demand requests for rate quotes through an agent.
- Ranks the best rates and providers and surfaces arbitrage opportunities and other trading metrics.
How I made it trustworthy
I iterated through human feedback and evaluation against real trading outcomes, and only then deployed it for fully autonomous operation. A rate that looks right but isn't costs money, so reliability came before speed.
Outcome
- Replaced 3+ hours a day of manual monitoring across dozens of group chats, so Treasury could spend that time on execution.
- Helped the team identify a key partner to work with, based on the system's output.
- Surfaced arbitrage opportunities the team acted on.
There's no dollar figure here. I don't see the team's realised trading results, so I won't guess at one.