Syvorai AI Lab

Specialized AI models

The Syvorai AI Lab trains, measures and publishes small models that solve real problems inside your infrastructure. The first of the family is Decisum, the PT-BR ticket classifier.

Meet Decisum See the benchmarks

Decisum
launched
2 × 322M · Apache-2.0
Ensemble of two 322M full-fine-tuned encoders that reads a PT-BR e-commerce ticket and answers category, decision and urgency in ~100 ms.
See the model
Veredicto
coming soon
9B · Apache-2.0
9B (clef-flash + LoRA), next in the line — coming soon.
Imperium
coming soon
26B · Apache-2.0
26B, the largest of the family — coming soon.

Measured on an honest ruler

Decisum v10+swa · overall score: 0.6339
Laya (zero-shot) · overall score: 0.2005
latency per case, on your GPU: ~100 ms

Overall score on the same gold ruler (n=753), never used in training.

Full benchmarks

November 10, 2026
A "System 1" Decision Model for PT-BR Customer Support — Part 4: The Judge Bake-off and the Hybrid v6
The judge bake-off against the human gold, the hybrid distillation of v6, the v7 to v12 experiments, the nine-challenger campaign, and the champion v10+swa ensemble published on Hugging Face under Apache 2.0.
November 3, 2026
A "System 1" Decision Model for PT-BR Customer Support — Part 3: The Ruler Was Wrong (Benchmark, Gold, and the v3→v4→v5 Trilogy)
The ruler was wrong: a paired benchmark against the zero-shot backbone, blind adjudication of 175 disagreements, a gold set of 1.000 tickets, and the v3 → v4 → v5 trilogy up to the root-cause fix — re-labeling the entire train set with a strong judge.
October 27, 2026
A "System 1" Decision Model for PT-BR Customer Support — Part 2: Results, Calibration, and the Verdict on RL
The results of decisor v2 and v3: gold set accuracy, P/R/F1 of the 17 categories, calibration via temperature scaling, three RLCD rounds with a negative verdict, and the audit that revealed the evaluation ruler was wrong.

Join the waitlist

Tell us about your use case and be notified when the next models in the family are published.

Count me in

Specialized AI models

The model family

Decisum

Veredicto

Imperium

Measured on an honest ruler

From the lab to the blog

A "System 1" Decision Model for PT-BR Customer Support — Part 4: The Judge Bake-off and the Hybrid v6

A "System 1" Decision Model for PT-BR Customer Support — Part 3: The Ruler Was Wrong (Benchmark, Gold, and the v3→v4→v5 Trilogy)

A "System 1" Decision Model for PT-BR Customer Support — Part 2: Results, Calibration, and the Verdict on RL

Join the waitlist