Scale
An autonomous bank operations agent that decides the right next step on each case, and proves it was safe to take. Built at the Real-World Agents Hackathon by Dwelly.
October 10, 2026
1 min read
Hackathon
AI Agents
Banking
Python
London
Scale, Real-World Agents Hackathon
Hosted by Dwelly in London. Banking and Financial Services track.
A suggestion is not permission. Scale reads a customer case the way a bank's back office gets it, the AI proposes one next step, and plain code checks it before anything runs.
Results
- 9 of 10 on our judgement tests: identity removed, hidden instructions, contradicted amounts, distress, role violations, ambiguous references
- Every case comes with the agent's reasoning written down
Stack
Python, LLM agents, our own eval harness with injected faults.
