Inspiration
LeakTwin — Project Story
LeakTwin started from a simple problem: teams keep adding SaaS and AI subscriptions, but most finance dashboards only show what was spent. They do not answer the harder question: what should we change next, how much will it save, and what is the risk?
LeakTwin is a Gemini-powered financial digital twin for software spend. It analyzes SaaS and AI subscription data, usage, seat counts, overlapping tools, and renewal timing. It then identifies recoverable spend, generates what-if scenarios, and turns the result into a prioritized recovery plan: Do today, Do this week, and Before renewal.
The goal was to move beyond a chatbot or static expense dashboard. LeakTwin is designed as an agentic workflow: input data is normalized, Gemini reasons across cost and usage signals, actions are ranked by savings and workflow risk, and approved recommendations can be converted into an execution plan.
The current prototype uses synthetic SaaS data so the demo is reproducible and privacy-safe. The architecture is designed to support invoices, screenshots, contracts, usage exports, and vendor APIs later.
LeakTwin uses Gemini 3.5 Flash through the Google GenAI SDK, with a FastAPI backend packaged for Google Cloud Run. The interface focuses on one outcome: showing how much money can be recovered and what should happen next.
The biggest lesson was that useful AI agents should not simply generate more text. They should make a defensible decision, quantify impact, surface uncertainty, and reduce the number of decisions the user has to make manually.
A key challenge was separating autonomous reasoning from irreversible actions. LeakTwin therefore separates decisioning from mutation: Gemini can analyze and propose actions, while high-impact vendor changes are designed to require explicit approval and scoped credentials.
The result is a system that turns software spend from passive reporting into an actionable recovery workflow.
Log in or sign up for Devpost to join the conversation.