Vroom Relief

Inspiration

Buying and financing a car is exciting, but managing the financial responsibilities that come afterward can quickly become overwhelming. Customers often struggle to understand their loan payments, refinancing options, vehicle equity, and what to do when their financial circumstances change. Finding clear answers can involve navigating complicated financial information or waiting to speak with a representative.

We were inspired to make automotive financial assistance more accessible, personalized, and conversational. Our goal was to build an AI-powered platform that doesn't just show customers their financial information but actually helps them understand it and make more informed decisions.

That's how Vroom Relief was born. We wanted to create an intelligent financial companion that brings vehicle information, financial insights, and human-like AI assistance together in one seamless experience.

What it does

Vroom Relief is an AI-powered automotive financial assistance platform that helps Toyota customers explore their vehicles, understand their financing, and discover their next best financial moves.

Our platform provides personalized dashboards where customers can view their loan information, monthly payments, estimated vehicle equity, and upcoming payment schedules. An interactive financial simulator allows customers to explore scenarios such as making additional principal payments, extending loan terms, and understanding potential interest savings.

At the center of our platform is Vroom AI, an intelligent assistant powered by NVIDIA Nemotron. Customers can ask questions about their financing, compare Toyota models, understand payment difficulties, and explore possible solutions through natural language.

We also integrated Retell AI to enable real-time voice conversations. Customers can speak directly with Vroom AI through their browser, while live captions display both the customer's speech and the AI's responses inside the chat interface. This creates a more natural and accessible experience for people who prefer speaking instead of typing.

Vroom Relief also supports specialist assistance requests for situations such as payment difficulties and trade-in consultations. To prioritize customer safety, sensitive actions use a two-step confirmation process, and the platform distinguishes financial estimates and pending requests from actual approvals or confirmed appointments.

Our current prototype uses simulated customer accounts to demonstrate how the experience adapts to different financial situations without requiring access to real customer financial records.

How we built it

We started by identifying common challenges faced by automotive financing customers and designing a platform that could connect financial information, vehicle exploration, and AI assistance.

For the frontend, we used React, TypeScript, and Vite to build a responsive Toyota-inspired interface with customer dashboards, vehicle comparisons, interactive financial charts, and an integrated AI assistant.

For the backend, we used Python, Django, and Daphne ASGI to manage authentication, customer information, financial calculations, API integrations, and WebSocket communication.

Our primary AI engine is NVIDIA Nemotron, which powers Vroom AI's reasoning and natural-language responses. Rather than allowing the model to generate financial figures independently, we connected it to backend calculation tools that produce consistent results based on the customer's simulated account.

For vehicle information, we integrated the NHTSA API and worked with CarImagesAPI to support vehicle data and imagery.

One of our biggest technical achievements was connecting Retell AI's Custom LLM agent directly to our existing Nemotron backend. This allowed us to use Retell for speech recognition and voice delivery while preserving Nemotron as the intelligence behind the conversation.

We also developed a live transcription system using WebSockets so customers could see their voice conversations appear inside the existing website chat.

Throughout development, we used Cursor and AI-assisted coding tools to accelerate implementation, debug integrations, and refine the user experience.

Challenges we ran into

One of our biggest challenges was integrating multiple AI services into a single reliable system.

Initially, our Retell voice assistant was using its own conversation-flow model instead of NVIDIA Nemotron. This meant the voice assistant and website chatbot were operating independently. We had to transition to a Custom LLM architecture and connect Retell to our Django backend through WebSockets.

We also faced issues with local server configuration. Our initial Django setup rejected WebSocket requests, resulting in connection errors and silent voice calls. We resolved this by configuring Daphne ASGI and exposing our local backend securely for testing through ngrok.

Another challenge involved frontend and backend synchronization. Even after fixing the voice integration, outdated frontend builds continued displaying configuration warnings. We had to trace the running frontend processes, remove obsolete connection logic, and ensure the browser was loading the latest implementation.

Making AI-generated financial responses trustworthy was another important challenge. We worked to prevent the assistant from exposing internal technical data, generating unsupported financial claims, or confusing a pending customer request with a completed action.

These challenges taught us that building a useful AI product requires much more than connecting an API. Every system needs to communicate reliably while protecting the customer's information.

Accomplishments that we're proud of

We are especially proud of successfully integrating NVIDIA Nemotron and Retell AI into one conversational experience. Our voice assistant can use the same backend financial tools as the website chatbot, providing consistent information across both interaction methods.

One successful test involved an extra-payment scenario where the voice assistant retrieved the same calculated result as the text assistant: approximately $961 in interest savings and six months saved by making an additional $100 monthly payment on a simulated account.

We are also proud of our live voice caption feature, which displays customer speech and AI responses inside the existing chat interface. This helps make the interaction more accessible and allows customers to follow along visually.

Another major accomplishment is our personalized financial dashboard. Instead of providing generic financial advice, Vroom Relief uses individual simulated customer situations to explain relevant options, potential tradeoffs, and next steps.

Finally, we're proud of building the platform around transparency and customer control. Our AI is designed to explain financial information rather than promise approvals or make unauthorized changes to financing agreements.

What we learned

This project gave us valuable experience building a full-stack application that combines real-time communication, AI reasoning, financial calculations, and external APIs.

We learned how to integrate NVIDIA Nemotron into a practical application, connect a custom language model to a voice platform, and manage WebSocket communication between multiple services.

We also gained a deeper understanding of the importance of authentication, customer-specific context, error handling, and reliable backend calculations when working with financial information.

One of our biggest takeaways was learning that AI should not simply provide an answer. It should be able to explain where that answer came from, recognize its limitations, and help users make informed decisions.

We also became more comfortable debugging complex integrations, working with frontend and backend technologies, and making architectural decisions under hackathon time constraints.

What's next for Vroom Relief

Our next step is to expand Vroom Relief from a working prototype into a more connected automotive financial assistance experience.

We want to integrate a real appointment scheduling system that allows customers to connect with appropriate representatives for refinancing questions, payment assistance, trade-ins, and general support. This includes verified representative availability, appointment confirmations, and email notifications.

We also plan to expand our telephone capabilities through Twilio and Retell AI, allowing customers to call Vroom AI from a traditional phone number and transition to human assistance when necessary.

Another major goal is implementing real-time synchronization through SpacetimeDB so that customer conversations, appointment requests, and important updates can remain consistent across web chat, browser voice, and phone interactions.

We hope to expand our vehicle information integrations, improve financial scenario modeling, and introduce more proactive recommendations based on changes in a customer's circumstances.

Ultimately, our vision is to make Vroom Relief a trusted AI-powered companion that helps people navigate automotive financing with greater clarity, confidence, and control.