Education & language
Speech, answers and translations checked by people who know the language.
Speech data in English and Indian languages, pronunciation scoring, learner answers graded to a rubric, and review of AI tutors.
From the first recording to the graded answer
Every part of the workflow where we can take work off your team.
Speech data
Learning
Language
AI tutors
We work with speech, learner answers, text and English and Indian languages.
What we label
Every class is written into a guide with your team before the first batch. Files come back as COCO JSON, YOLO, Pascal VOC, CSV, JSONL or your own schema.
Speech
Transcripts
Time span and text
Verbatim words, fillers, noise
Pronunciation
Time span and class
Correct, close, wrong, by sound
Speakers
Time span and class
Speaker, language, accent
Text
Learner answers
Rubric
Score, reason, error type
Translations
Rubric and span
Meaning kept, meaning lost, fluency
How we test the model
The tests that matter for education & language models, and what you get back from each.
Grading agreement
Answers scored by two reviewers and the model, to see where they differ.
You get backAgreement by question and score band
Speech recognition by language
Transcripts checked against the audio for each language and accent.
You get backWord error rate by language and accent
Tutor answer quality
AI tutor answers scored for accuracy, tone and level.
You get backScores by subject, with failing examples
Where education & language teams start
Problems teams bring us, and the first piece of work we would scope.
Your speech model works in English and struggles in Indian languages.
What we doSpeech collected and transcribed across languages and accents, with code switching marked.CollectAutomated grading disagrees with your teachers.
What we doAnswers graded to your rubric by trained reviewers, with the reason for every score.AnnotateYour AI tutor sometimes gives wrong or confusing answers.
What we doTutor answers reviewed against a rubric before and after each release.Evaluate & alignRelated work
Client and volumes [REAL DATA: Client name and volumes]
Bring us the education & language data your model gets wrong.
Send a small batch and get it back done, free, usually within 24 to 48 hours. You judge the work before you pay for anything.