AutoValue AI demo walkthrough

Recorded October 1, 2026. This silent, captioned recording follows real interactions with the local application and its verified RF05 model. Estimates reflect historical 2023 U.S. advertised asking prices. The River segment replays synthetic research results. AutoValue AI remains in active development.

1. 00:00

   AutoValue AI. A working prototype for historical 2023 U.S. asking-price estimates. Recorded against the local API.

2. 00:05

   Enter a vehicle. Select a 2020 Toyota Camry with 48,000 miles and a 90% prediction interval.

3. 00:11

   Estimate with RF05. The API returns a price estimate and a calibrated range. These are historical asking prices.

4. 00:19

   Compare interval coverage. Choose 95% coverage: the interval widens while the point estimate stays the same.

5. 00:29

   Try another vehicle. Estimate a certified 2021 Honda CR-V with 31,000 miles, using 90% coverage.

6. 00:38

   Review saved estimates. The API saves these requests in browser-scoped SQLite history. Three estimates are now listed.

7. 00:45

   Inspect the model. The engineering view identifies the frozen RF05 model and checks that its serving artifacts are ready.

8. 00:51

   Read the holdout results. On 27,589 held-out observations, mean absolute error is $10,575. The model has material limitations.

9. 01:00

   Check interval calibration. The nominal 90% interval reached 89.10% empirical coverage. Wider intervals carry a cost in precision.

10. 01:08

    Follow the architecture. RF05 serves the estimate. River learning and external-source experiments run in separate paths.

11. 01:16

    Review experiment decisions. The experiment log records accepted, rejected, and research-only results.

12. 01:23

    Replay a synthetic shift. Replay 600 simulated mileage-shift events. This shows recorded research results, separate from the serving model.