We're Live on Product Hunt — Forward-Looking Used-Car Price Forecasts
Today is launch day. CarCast is live on Product Hunt and we'd love your support: carcast.ai on Product Hunt →
Here's the short version of why we built this and what makes it different.
Why we built CarCast
Every used-car pricing tool a dealer uses today — Manheim Market Report, KBB, Black Book, vAuto — answers the same question: what did this vehicle sell for last week? They're backward-looking averages, and in a stable market that's enough.
But the used-car market in 2024–2026 has been anything but stable. Wholesale truck prices were up 12% in three months, then corrected hard. EV residuals collapsed. Tax-season demand pushed mid-size SUVs through the roof. Dealers were making six-figure inventory bets using yesterday's data and the data was already wrong by the time they paid.
CarCast answers a different question: where are prices heading?
What we actually do
We forecast used-car prices 30 and 60 days out across 300+ vehicle segments — from a 2022 Ford F-150 XLT to a 2020 Tesla Model Y Long Range to a 1972 Chevrolet C-10. Each forecast comes with:
- P10 / P50 / P90 confidence bands — not a single number, a range with probabilities
- A trend classification — Rising / Stable / Softening
- Macro context — the Manheim Used Vehicle Value Index and the BLS used-car CPI alongside every segment
The forecasts re-run every Sunday morning against fresh dealer-listing data. The hub at /market-trends shows the full national picture, refreshed daily.
The technology, briefly
The forecasting stack is a 200M-parameter foundation model for time-series, fed weekly with:
- 52+ weeks of dealer-listing prices (median, average, P10, P90)
- The Manheim Used Vehicle Value Index (wholesale benchmark)
- The BLS used-car CPI (retail benchmark)
- Per-segment listing depth and average mileage
We don't predict next week's price; we produce a probability distribution over the next 8 weeks. Confidence bands matter more than point estimates when you're making a buy/sell decision.
The infrastructure is FastAPI + Postgres on Fly.io, a Next.js front-end on Vercel, and Modal-hosted GPU inference for the model. The forecast worker is a single APScheduler cron that batches all 300+ segments every Sunday at 3 AM ET.
Who uses CarCast
- Independent used-car dealers timing auction buys and inventory turn
- Wholesalers spotting segment-level shifts before the broader market does
- Fleet remarketing teams forecasting end-of-cycle residuals
- Retail buyers deciding whether to buy now or wait
Free plan covers 5 segments. Pro at $149/mo unlocks all 300+ segments and the 60-day horizon. Agency at $499/mo adds API access for dealers integrating into a DMS.
What we'd love your help with
If you're on Product Hunt today:
- Try the free plan — pick the 5 vehicles you care about and tell us where the model is wrong
- Leave a comment with the vehicle segment you wish we covered
- Roast the product — what's missing, what's confusing, what doesn't make sense
We'll be reading and replying all day.
— Rick
Forecasts and trend classifications are informational analytics only and do not constitute financial advice or a recommendation to purchase, hold, or sell any vehicle.