Status
Model tuning
Project
CEFR.ai
Language difficulty modelling
Builds interpretable CEFR-level predictors from lexical, syntactic, and semantic features.
Dataset size
96k labeled passages
Last update
2026-03-08
Objective
Ship adaptive language products with measurable difficulty calibration and feedback loops.
Revenue Metrics
Monthly revenue
$1,280.00
Direct spend
$18.00
Total spend
$44.08
Net monthly
$1,235.92
Datasets
CEFR-Aligned Text Benchmark
Multi-domain passages with CEFR labels, readability metrics, and metadata.
96,340 records
Daily refresh
Tuning labels
2026-03-08
Recent Experiments
2026-03-08
Prompt-augmented CEFR classification on noisy learner text
Reduced borderline A2/B1 misclassification by 12%.
Open code ->2026-03-03
Lexical rarity feature ablation for CEFR bands
Feature had high impact on B2/C1 separation; kept in stack.