PiggyBank

Money that talks

Comment 9

PiggyBank: Make your money talk!

💡 Inspiration: The Invisible Spending Problem

We built PiggyBank because personal finance has quietly become one of the biggest sources of everyday stress.

Most people don’t struggle because of big, one-time purchases. They struggle because of the tiny, invisible habits — a $6 coffee here, a $12 delivery there — that quietly drain financial confidence and compound stress over time.

The Financial Anxiety Data

  • 65% of Americans live paycheck to paycheck. (LendingClub, 2024)
  • 77% of adults feel anxious about money. (APA, 2023)
  • 59% of Gen Z and Millennials admit to impulse purchases. (Credit Karma)

🎯 Use Case: Awareness in the Moment

We built Piggy to do more than just track spending — it’s an agent quietly working in the background, learning your financial habits and helping you make smarter choices.

Imagine this: you tap your card at Starbucks, and Piggy takes note. Over time, it learns that you grab coffee every day around 8 a.m. Since Piggy is designed to understand your behavioral patterns, it predicts when your next purchase is coming — and before you buy(say at 7.30), it sends a friendly iMessage nudge:

“Hey, maybe try a cheaper alternative like Wawa today? You ould be a day closer to your new bike that you are saving for.”

You reply, and Piggy keeps learning. Soon, it’s not just tracking your spending — it’s helping you visualize how small choices add up. Instead of cutting out your morning coffee entirely, Piggy helps you see how redirecting those savings could make your new $250 bike a reality faster than you thought.

Piggy doesn’t guilt-trip you — it guides you, helping you reflect at the moment when awareness matters most.

Also, to reduce user friction, if you do an in-person purchase AKA have receipts, just text them to piggy, and he'll handle the rest! Super Simple!

✨ What It Does: A Real-Time Financial Conscience

PiggyBank is a real-time spending coach powered by a feedback loop that connects your transactions to a conversational agent.

Core Functionality

Item-Level Transaction Ingestion: Uses the Knot ** Production **TransactionLink API to ingest granular, item-level purchase data into Snowflake securely.

Real-Time Classification & Feedback: Each purchase triggers a real-time loop:

  1. The backend uses Dedalus to classify the item (Need/Want).
  2. Sends a personalized message via Photon to your iMessage for quick feedback.
  3. Your reply fine-tunes Piggy’s predictions.

Proactive Prediction: Detects recurring spends and sends nudges before they happen.

Multimodal Receipt Processing: Powered by Gemini 2.5 Flash — users can snap a photo of a receipt, and Piggy instantly extracts structured data via OCR.

“Oink oink! I’ve analyzed your receipt from Courtyard by Marriott. Total: $23.67. Saved to your tracker!”

Contextual Chat Assistant: A GPT-4–powered chat assistant that reads live Snowflake history to answer:

“How much did I spend on coffee this month?” or “What could I cut to save $50 next week?”

🧠 How Our Algorithm is Made (Deployed on DigitalOcean)

Our predictive and conversational intelligence layer runs on a modular, serverless architecture using DigitalOcean Gradient AI.

1. Data Foundation & Feature Store (Snowflake)

All transaction data and user feedback are stored in Snowflake, serving as the feature store where SQL queries and vector search aggregate behavioral patterns.

2. Predictive Behavior Model (DigitalOcean Gradient AI)

Llama 3.1 is deployed to intelligently ingest these datapoints and predict when the next purchase is likely to occur, allowing the agent to proactively trigger timely recommendations. With the help of Python data processing and Snowflake time-series model analyzes each user’s purchase history, computes typical intervals between repeat buys, and produces a numeric next-purchase timestamp plus confidence. Piggy analyzes factors such as the size of previous orders (e.g., whether a grocery trip typically covers a week or a month) and historical spending frequency to forecast upcoming transactions with high accuracy.

3. Classification & Logic (Dedalus Labs)

Dedalus handles initial transaction classification and confidence scoring before engaging users via Photon.

4. Multimodal Processing (Gemini API)

Receipt image processing via Gemini 2.5 Flash enables OCR and JSON extraction before saving structured data to Snowflake.

🛠️ Tech Stack

Layer Technologies Key Use
Backend / API FastAPI (Python), Flask, Node.js Microservices, RESTful endpoints, Core Analytics
Data Warehouse Snowflake Central time-series storage and analytics
Integrations Knot API, Clerk Auth Secure item-level data ingestion, Auth
Conversational AI Photon, iMessage SDK Real-time chat, proactive alerts
Intelligence Gemini 2.5 Flash, Dedalus, GPT4, Llama 3.1, DO Gradient AI OCR, Classification, Prediction
Frontend React (Vite), Clerk Auth Dashboard and visualization

🏆 Special Tracks and Prizes

Prize Name what we did for every track
Best Practical AI Innovation by Amazon Practical AI agent that improves real-world financial outcomes.
Best Financial Hack by Capital One Provides actionable coaching that translates abstract numbers into savings goals.
Build on the Knot TransactionLink API by Knot API Uses Knot for secure, item-level data granularity.
Exploring Hybrid Intelligence by Photon True Hybrid Agent integrating Node.js front-end and Python backend.
Best Use of Dedalus by Dedalus Labs Powers classification and confidence scoring for precise messaging.
Best Predictive Intelligence by Chestnut Forty Forecasts future spends to power core agent functionality.
Best Use of Gemini API by MLH Uses Gemini 2.5 Flash for accurate OCR-based receipt parsing.
Best Use of DigitalOcean Gradient AI by MLH Deploys predictive model on Gradient AI platform.
Best Use of Snowflake API by MLH Snowflake serves as the high-performance data backbone.

🚀 Accomplishments We’re Proud Of

  • Achieved a fully functional MVP with real-time classification, messaging, and prediction.
  • Developed a predictive commerce layer that forecasts recurring purchases.
  • Integrated a dynamic multimodal conversational interface (Photon + Gemini).
  • Built a scalable architecture ready for fintech deployment.

⏭️ What’s Next: From Nudger to Navigator

Piggy will evolve from a coach into a proactive financial navigator:

  • Foresight Meets Automation: Detect savings patterns and reallocate funds automatically.
  • Money Duolingo: Gamifies savings habits through streaks and literacy levels.
  • Embeddable Coach: Offered as a plug-in for banks and fintechs via secure APIs.

oink oink!

Built With

+ 9 more
Share this project:

Updates