Own your intelligence.

01 / THE CHALLENGE

Your idea. Your model.

Build something useful, ambitious, or unexpected with a custom model. Bring an agent, a specialist, a creative tool, or an idea of your own. Choose the approach that fits your project.

Start with the River documentation and talk to the River team about model access and your idea.

02 / AN OPTIONAL EXAMPLE

A model that learns your writing style.

The style-chat example learns from your conversation. Every eight user messages trigger a LoRA training step; the latest loaded adapter rewrites the assistant’s replies. Use it for inspiration or build something entirely different.

For this example, you’ll need Python 3.12+, uv, and a River API key. Save both files in the same folder.

export RIVER_API_KEY="your-river-api-key"
uv run --no-project style_chat.py

Use /status to check progress and /quit to finish. On Windows, set the key with $env:RIVER_API_KEY="your-river-api-key" in PowerShell.

Check the example: run uv run --no-project test_style_chat.py -v. Seven unit tests check batching, checkpoint restoration, and adapter loading with simulated training. They need no API key and do not verify live training or model quality.

03 / SHOW WHAT YOU BUILT

Make the custom model matter.

• Demo the experience. What did you build, and who is it for?
• Explain what the model learned. Show your training examples or reward signal.
• Show why it helps. Compare results on unseen tasks and share what you learned.

Projects are due at 5:00 PM Pacific. Follow the organizers’ submission instructions. See the YC event page for registration and the current schedule.