SpendWise AI

An autonomous, AI-driven personal financial coach that plugs into PayPal to predict cash-flow deficits, categorize live sandbox transactions, and execute proactive budget optimizations

Comment

Inspiration

Traditional personal finance tools are broken because they are completely retroactive. They act as digital autopsy reports—showing you colorful pie charts of money you have already lost after the damage is done. We wanted to flip this paradigm on its head. The rise of autonomous AI systems gave us a clear vision: a financial assistant shouldn't just send you a passive notification when you are broke; it should look ahead, predict the crisis, and provide the secure transactional tools to fix it before it happens. This drove us to build SpendWise AI.

What it does

SpendWise AI transforms personal budgeting from a passive chore into an active, protective financial relationship. By securely pulling transaction feeds into our engine, the platform delivers three primary features:

  • Predictive Cash-Flow Forecasting: Instead of simple history graphs, our engine runs transactional arrays through advanced language models to project cash balances weeks in advance and flag upcoming subscription overshoots.
  • Proactive AI Coaching Banners: When the model discovers an impending deficit, the assistant populates the user interface with distinct, actionable choices (e.g., budget reallocations or identifying redundant recurring bills).
  • Instant PayPal Remediation: Users don't have to leave the tab to fix their budget. They can click "Fix via PayPal" right inside the AI recommendation box to trigger simulated sandbox transfer optimizations instantly.

How I built it

To move from an idea to a fully working prototype in a short hackathon window, I utilized a lean, powerful stack of modern free-tier development tools:

  • Frontend Design & UI: Generated using Lovable.dev and Bolt.new to rapidly assemble a highly responsive, emerald-accented fintech dashboard using React, Tailwind CSS, and Lucide icons.
  • Database & Auth Architecture: Handled entirely by Supabase (PostgreSQL) to manage secure user sessions and store transaction arrays.
  • Backend Pipeline Orchestration: Visualized and deployed via Buildship, allowing me to build an active API webhook endpoint connecting my database to the AI models without writing a standalone Express server.
  • AI Intelligence Core: Powered by Gemini 1.5 Flash / Claude 3.5 Sonnet to evaluate raw transaction arrays and output structured data.
  • Transaction Engine: Directly integrated with the PayPal Developer Sandbox API to test live checkout validation, authorization loops, and mock account transfers.

Challenges I ran into

The most frustrating roadblock was forcing the LLM to return strictly deterministic outputs. In early builds, the AI would include polite conversational filler phrases alongside its financial tracking analysis, which immediately caused the frontend JSON parsers to crash. I solved this by writing strict system prompt guardrails that force the model to output a minified JSON schema with zero external text. Additionally, chaining dynamic Supabase row queries directly into chronological arrays inside Buildship required intense optimization of our backend filtration parameters to keep response times low.

Accomplishments that I'm proud of

I am incredibly proud of building a fully functional loop that bridges autonomous AI analytics with live transaction actions. Moving past basic text summaries and building a button that actually hits the PayPal Sandbox API endpoints to update user accounts was a huge milestone. Furthermore, taking the project from a blank screen to a deeply integrated, connected full-stack application within the hackathon timeline feels like a massive win.

What I learned

This project opened my eyes to the incredible power of agentic commerce—proving that AI shouldn't just be used to chat, but to actively perform helpful work on behalf of users. On a technical level, I learned the massive value of visual workflow builders like Buildship to skip repetitive API setup work, allowing me to focus 100% of my energy on crafting core business logic and refining the product's value proposition.

What's next for SpendWise AI

The immediate next step is expanding our PayPal implementation from basic Sandbox transfers to utilizing PayPal Subscriptions and Smart Invoicing APIs, allowing the AI to automatically negotiate or pause active billings on behalf of the user. We also plan to integrate true multi-agent systems where separate AI nodes can securely run competitive comparisons on commercial utilities to find cheaper alternative plans for the user automatically.

Built With

Share this project:

Updates