Hi Product Hunt 👋 I’m Julien, maker of SelfJev.

I love decision models like TypeSafe’s Jev.

Instead of asking an LLM to write an answer and then parsing it, you ask a typed question like:

→ “Does the customer want a refund?”
→ “Which team should handle this?”

…and get a calibrated probability back.

They’re fast, cheap, and don’t hallucinate output formats.

The problem: some teams can’t send sensitive text to someone else’s API — support tickets, medical notes, contracts, internal data.

So I built SelfJev: an open decision model you run on your own GPU.

4 answer types

🔌 Jev-compatible API

Already using TypeSafe’s Python SDK?

Just point:

TYPESAFE_BASE_URL
TYPESAFE_API_KEY

to your SelfJev server.

No application code changes.

🌳 Shared computation

SelfJev reads the document once, then reuses that computation across all questions and answer choices using a shared-prefix tree.

So asking many questions about the same document costs little more than asking one.

🎯 Fine-tuning

Fine-tune on your own labels with:

selfjev finetune

Or use the OpenAI-style HTTP fine-tuning API.

🚀 Self-hosting

Run locally:

pip install "selfjev[serve]" && selfjev serve

Works on a 24 GB GPU.

Or deploy to AWS with:

selfjev deploy aws up

On our evaluation suites:

These are our own suites with AI-authored labels, not a universal benchmark or ranking.

Every evaluation report is available in the repo, and the dataset is public on Hugging Face as Decision Bench.

It’s not a hosted service.

There’s no SelfJev cloud.

You run the model. Your data stays with you.

The code is Apache-2.0, and I’ve also made the full research journal public — including the experiments and dead ends.

I’d love to hear what decisions you’d use SelfJev for, and especially where it gets them wrong. 🙂
model -> https://huggingface.co/Jwuthrich