Inspiration
Most people sign car paperwork without really understanding it. APR, residual value, lease equity: the words sound complicated, and confusion is exactly when people make expensive mistakes. At HackUTD 2025 I built a Toyota app with a Learn section for financial literacy, and the judges asked me to pitch it to Toyota. This year I wanted to go further. Instead of another chatbot you have to figure out, what if Toyota just called you, walked you through your options like a friend who's good with money, and made sure you actually understood them?
What it does
- You tap "Get a call." Your phone rings from a real Texas number within seconds.
- It finds out who you are by voice. Current customers say their customer code and the AI pulls up their car and loan or lease. New customers get a new customer ID on the spot.
- It asks only what matters: what changed in your life, how much you drive, what you can afford, and what you care about most.
- Your screen fills in while you talk. Your answers appear as you give them, and the plan pops up mid-call: keep the car, pay it off sooner, refinance, trade up, lease or certified pre-owned, each with an exact monthly payment.
- It teaches while it helps. Every option shows how the number was calculated and the trade-offs in plain English. Ask "what does residual mean?" mid-call and it explains in one sentence with an everyday example, then saves that word to your Learn section.
- Learn section: short lessons on APR, interest, down payments, credit scores, leases and total cost of ownership. Each one has a real-number example (like how a few points of APR can cost thousands more on the same car), rules of thumb like 20/4/10, and a quick quiz.
- The whole conversation is saved like a chat, along with a written recap of your plan, your next step and your customer ID.
- A real person is one step away. If money is tight, the AI does not upsell. It hands you to a Toyota specialist, who sees the full conversation in the Toyota staff view.
How we built it
- Voice and phone: an ElevenLabs voice agent running a Claude model, calling out through a Photon SIP phone number.
- Tools, not guesses: during the call the agent uses seven tools on my backend: identify the caller, look up the account, record each answer, calculate options, explain a term, hand off to a person, and wrap up the call. Every payment comes from a finance engine (amortization, lease math, Texas sales tax), never from the language model.
- Plain-language glossary: the "explain a term" tool pulls from a glossary of car finance words, each with a one-line definition, an everyday analogy and why it matters, so explanations stay accurate and consistent.
- Live screen: the backend pushes each answer and each line of the conversation to the browser over Socket.IO.
- Frontend: React, TypeScript and Vite with Framer Motion animations. Lesson examples in Learn are calculated with the same finance math, so the numbers are real.
- Hosting: one Node.js service on Render that serves the site and points the voice agent at itself on startup.
- Demo data: four made-up customers with only a name, a customer code and their car and contract. Everything else has to come from the call.
Challenges we ran into
- Keeping the AI honest about money. The fix was architectural: the model can only quote numbers a tool returns.
- Explaining finance without dumbing it down. Every term needed a definition that is short enough to say on a phone call but still correct.
- Calls start anonymous, so the agent identifies people by voice. Spoken codes like "T M seven three nine zero" are cleaned up before lookup, and anyone without a code becomes a new customer instead of a dead end.
- Getting the conversation onto the screen and knowing exactly when the call hangs up.
- Text delivery: US carriers require business numbers to be registered before they deliver texts, which takes days, so during the hackathon the recap shows in the app.
- Building it solo in one weekend.
Accomplishments that we're proud of
- Completing this project in its entirety, solo.
- A real phone call that ends with a real plan you actually understand, with numbers you can check, in about three minutes.
- Financial literacy built into the product, not hidden in a help page: every number shows its work, every term can be explained mid-call, and every lesson ends with a quiz.
- Responsible by design: it says it's an AI up front, asks for consent before calling, never promises an approval, and sends hardship to a person instead of an upsell.
- It works for first-time buyers who aren't Toyota customers yet, not just existing accounts.
What we learned
People don't need more options. They need to understand the ones in front of them. A plain explanation at the right moment builds more trust than any amount of polish, and voice AI only feels magical when the boring parts are solid: knowing who is calling, tools that never fail, and showing your work.
What's next for The Fast and the Financed
- Connect it to real Toyota Financial Services accounts and rate tables.
- Track which terms customers ask about most and turn them into new lessons.
- Call customers before a lease ends or when they could save by refinancing.
- Finish carrier registration so the recap arrives by text.
- Spanish and other languages, so more people can understand their own car deal.
Built With
- anthropic
- claude
- css3
- elevenlabs
- express.js
- frame-motion
- html5
- javascript
- node.js
- photon
- react
- render
- sip
- socket.io
- typescript
- vite
- vitest
