CarTender

A snapshot of loan financing options
A snapshot of vehicle history & safety
User's purchasing choice
Financial services for potential buyers
Cars we swipe left/right on
Financing summary for chosen car
Like page for all cars you swipe right on

Inspiration

In a generation where people find the love of their lives on an app, we seek to get users in their dream cars in a similar way!

What it does

CarTender uses modern agent frameworks as well as personalization to find the best car for YOU.

Answer a few simple questions (mostly about your finances and some car must-haves) to get started.
CarTender's agents then go to work: aggregating reviews, listings, rankings and other info online to provide you suggestions that you could swipe left or right on like a dating app (CarTinder??)

Agents are also used to guide you through the purchasing process, providing the best rates, insurance quotes, maintenance and inspection plans.

How we built it

Vue.js frontend, Flask backend
LangGraph agents running on NVIDIA-nemotron-nano-9b-v2
Compute, ML and LLM hosted on NVIDIA L4 24 gb instance on Brev.
Numerous APIs for car info, review, sentiment, financial, etc. information

Challenges we ran into

Designing semantic embeddings and personalization pipelines that capture subtle preferences (e.g., “no American cars,” “must have CarPlay”)
Balancing agent complexity with latency - LLM reasoning can be slow
Managing hosted GPU compute on Brev vs. simpler serverless architectures

Accomplishments that we're proud of

A fully agentic end to end car matchmaking platform in 24 hours
Multi agent orchestration
Submitting :)

What we learned

How to host and tune large models like Nemotron efficiently on GPU instances
Practical challenges of balancing response time, caching, and reasoning depth in multi-agent systems
How to combine structured APIs with unstructured LLM reasoning for real-world decision making