Wallet Guard

geofencing-info-slide
mapped-ui

Inspiration

Most banking and budgeting apps are fundamentally flawed, they are entirely reactive. They show you a ledger of what you've already spent, effectively telling you that you are broke after the damage is done. We realized that true financial protection isn't about tracking past mistakes, it's about intercepting them.

We wanted to build something that proactively enforces good financial habits by introducing healthy friction exactly when and where it matters most: at the point of sale. Wallet Warden was inspired by the idea of an "overly attached" financial bodyguard an AI that knows your budget, and actively intervenes to stop impulse buys before you even reach the checkout counter.

It dwells on the 21 day habit method to identify your financially weak points and use smart contracts to reward you for healthy financial habits.

How we built it

Wallet Warden is a fusion of spatial intelligence, Web3 infrastructure, and conversational AI. Here is how we orchestrated the stack:

The Spatial Trigger (Radar.io): We set up geofence triggers around high risk areas based on users spending patterns . When a user physically crosses this boundary, Radar.io fires a webhook to our backend.

The Brain (Python Backend & Nessie API): Our backend catches the webhook and immediately queries the Nessie API to check the user's available balance and specific category budget (e.g., fetching the "Electronics" budget when entering a Best Buy).

The Web3 Ledger (Solana): Instead of just logging transactions, we use blockchain as an immutable commitment device. Users who are in an "active planning state" can sign up for a 21-day financial achievement challenge. They voluntarily lock a set amount of their savings into a Solana smart contract. The blockchain ensures strict accountability, acting as an untouchable vault that only releases the funds once the 21-day habit-building period is successfully completed.

The Intercept (Twilio & ElevenLabs): We combined Twilio with a conversational Voice AI agent. The backend injects the real-time Nessie data into the AI's system prompt and triggers an outbound phone call to the user. The AI acts as a strict financial concierge, reminding them of their budget, actively talking them out of impulse buys, and even utilizing function calling to scrape the web for price matches at competing stores.

The Time Machine (Nano Banana 2): To keep users motivated during their 21-day lock, we built a "Time Machine" dashboard. We map out their financial progress using dynamic data graphs that project their future savings trajectory. To make this data emotionally resonant, we integrated the Nano Banana 2 model to generate personalized, aspirational images of the user's future, visually bringing their financial goals to life right next to their progress charts.

Challenges we ran into

Our most complex hurdle was accurately predicting user intent. Specifically, determining if a user was actually entering a store to make a purchase or simply walking past it. Relying purely on a broad geofence initially led to frustrating false positives, where the AI would call a user who was just walking down the sidewalk outside of a retail zone. We had to heavily refine our spatial logic, experimenting with dwell times and tighter radius precision, to ensure the AI intervention only triggered when the user was genuinely inside and lingering.

Stringing together a webhook, fetching banking data, interacting with a Solana RPC node, and spinning up an ElevenLabs-powered voice call required us to aggressively optimize our backend architecture. Additionally, preventing the Voice AI from hallucinating financial numbers required strict prompt engineering and locking down the agent's context window exclusively to the data provided by the Nessie API.

What we learned

We learned the massive potential of combining real-time function calling with voice agents. Seeing an AI successfully negotiate a user to not make an unnecessary purchase.

We also learned how to successfully bridge the gap between physical, real-world actions (geofencing) and decentralized state management (Solana). Most importantly, we realized that forcing a moment of proactive friction, a literal phone call to break your shopping momentum is an incredibly effective psychological tool for building better financial habits.