Hello Product Hunt! 👋

I’m Hanna, one of the creators of Arkor.

We built Arkor because we keep running into the same gap: adding an AI feature to an application is easy, but adapting a model for a specific task still feels like joining an ML infrastructure team.

When you want to fine-tune, you often maintain a separate Python project, convert datasets to unfamiliar formats, provision GPUs, and move between scripts, notebooks, and dashboards.

We wanted model development to feel like normal software development.

With Arkor, the workflow starts within your existing repository:

1. Tell the cloud code or codecs how you want the model to behave.

2. Your coding agent can search or generate datasets, write conversion scripts, create TypeScript trainers, and add diagnostics.

3. Review generated code and changes.

4. The agent runs pnpm dev, and Arkor Studio opens locally at localhost:4000.

5. Click Run the training.monitor damage and checkpoints, test the trained adapter, and deploy the result.

For example, you can give your coding agent prompts like:

Use Arkor to fix a model that rewrites rough drafts as tweets in my style: https://github.com/arkorlab/arkor
Find or generate a suitable dataset, create a TypeScript training workflow, and let me know when it’s ready for review in Arkor Studio.

Arcore isn’t meant to be a magical prompt-to-model black box. It’s a developer-controlled workflow where coding agents can handle most of the setup, while training code, data transformation, evaluation, and final decisions remain visible and reviewable.

We are particularly interested in feedback on:

• Does the Coding Agent workflow feel intuitive?

• What parts of fine-tuning still feel unclear or scary?

• What you’ll need before using Arcore for a production model.

• What models, datasets, and deployment workflows should we support next?

Thanks for checking out Arkor. We’ll be here during the launch and would love to hear what you try to build. 🙏