$42 Billion. Gone in 24 hours.
That wasn’t a glitch—that was the reality of the 2023 banking crisis. Today, financial panic moves at the speed of a trending tweet, yet the risk managers protecting our global banks are often still relying on slow, static Excel spreadsheets.
When the H0 Hackathon by Vercel and AWS was announced, my teammates (Sahil Kumar and Girija Unde) and I decided to tackle this exact problem. As a Computer Science student at Banasthali Vidyapith diving deep into AI and Machine Learning, I wanted to see if we could bridge the gap between predictive machine learning and enterprise-grade cloud architecture.
The result? LiquiShield—an AI-powered early-warning command center built to predict and visualize liquidity crises before they hit the news cycle.
Here is a deep dive into how we engineered the future of finance over a single hackathon weekend.
🏗️ The Architecture: Hacking the Future of Finance
To build a real-time stress-testing sideport, we needed a stack that could handle natural language processing, heavy time-series forecasting, and massive database scaling without breaking a sweat.
The "Panic Pulse" (NLP Engine)
Fear has a frequency, and we needed to measure it. We implemented a custom NLP model to analyze the mood of macroeconomic headlines. Instead of managers guessing how a news story might affect the market, our AI reads the text, calculates a "Crisis Vector," and translates that into a quantifiable market fear metric.The Crystal Ball (Facebook Prophet)
We wanted to provide hard, actionable data, not just vague warnings. We integrated Facebook Prophet to handle our time-series forecasting. We built a simulator with a custom "Severity Multiplier" and adjustable Runoff Rates (Retail vs. Corporate). When a user tweaks these sliders, Prophet instantly generates a 30-day cash depletion forecast.The Infinity Gauntlet (AWS Aurora Serverless)
This was the most critical piece of the puzzle. When a risk manager hits "Run Stress Simulation," thousands of relational data points need to be processed simultaneously to update the global branch network.
We strapped our backend onto Amazon Aurora PostgreSQL Serverless.
Why Serverless? When a crisis hits and the simulation runs, Aurora instantly scales up its compute capacity in milliseconds to handle the heavy math without the application freezing or lagging.
Cost-Efficiency: Once the simulation is over, it dynamically scales back down, making it an incredibly efficient architecture for unpredictable workloads.
🚧 The Challenges We Faced
Building an enterprise-level tool over a weekend is exactly as chaotic as it sounds. We dealt with 2 AM error messages that made absolutely no sense and deployment crashes that almost broke our spirits.
Getting Facebook Prophet to play nicely with our custom UI sliders required some intense debugging, and configuring the AWS Aurora backend to communicate seamlessly with our frontend under load was a massive learning curve. But fueled by coffee and pure stubbornness, we pulled it off and got to see our Zero-Hour Alarm successfully trigger in our live demo.
🚀 See It In Action
Check out the madness for yourself!
📂 Code: https://lnkd.in/dzj26Qk6
🎥 Demo: https://lnkd.in/d4QuwsvH
🌐 Live Site: https://lnkd.in/drs_bBX3
If you are a developer interested in FinTech, time-series forecasting, or AWS architecture, I’d love to hear your thoughts in the comments below! #FinTech #H0Hackathon #AWS #MachineLearning #AI #Vercel #SoftwareEngineering #BankingInnovation
Top comments (0)