Service
Ler em portuguêsPortuguese dataset and fine-tuning
Curated, documented and evaluated data to adapt models to Portuguese and the company domain.
Observed problem
A generic model does not master the domain language, format or constraints, while existing data contains duplicates, noise or personal data.
Expected outcome
A versioned dataset and LoRA-adapted model, both documented with a frozen evaluation split and baseline comparison.
What the scope includes.
- controlled collection or generation, cleaning and deduplication
- personal-data removal and train/validation/test splits
- LoRA fine-tuning, evaluation and private or Hugging Face delivery
How we begin
Projects scoped in USD or EUR
A discovery conversation defines the bottleneck, dependencies, evidence of completion and what stays out. Pricing follows that scope.
Engagements can be structured in USD, EUR or BRL. Timelines and pricing depend on validated scope.
Discuss a project