Our commitments.
They're who we are.
Ayni is a voice-native visual thinking platform for K-12. We are building it because too many students think faster than they can write, and the schools serving them deserve tools that listen first.
Universal design is at the center of how we build. The page below is the working set of commitments we hold ourselves to, the frameworks we align with, and the things we will not do. Click any card to read more.
The non-negotiables
Speech-to-text runs on the student's device. The audio never leaves the machine, and it is discarded the moment text is produced. The platform does not store voice samples, generate voice prints, or perform biometric analysis of any kind.
Every signal we surface to teachers comes from text. That choice is permanent. It is the foundation that lets us claim BIPA, FERPA, and COPPA defensibility by design rather than by policy.
Nothing a student writes or says inside Ayni is used to train, fine-tune, or improve any model the company runs or any model offered by our vendors. Contracts with school districts will state this in plain language.
There is no free or freemium tier. Free tiers in this category usually pay for themselves in student data. We refused that trade at the start so we never have to negotiate it later.
The animated write-in is the product. A student speaks and watches their thinking appear in front of them. AI features are user-triggered, never autoplaying. The suggestion bubble offers itself once and does not nag.
Blank slate means no suggestions. We respect that choice. The default posture of the system is to listen and reflect, not to recommend.
Be clear-eyed about this. What teachers see in Ayni will inform IEP goals, instructional moves, and conversations with families. That isn't a side effect of the product. It's the point of making thinking visible.
What the product will not do is make those decisions itself. Ayni does not generate placement recommendations, behavior scores, risk flags, or automated reports for administrators. It does not rank or classify students. The judgment belongs to the educator and the IEP team. We make thinking visible. Interpretation is human work.
The thinking architecture inside Ayni is built on decades of self-regulated learning research. The Start, Notice, Reflect, Connect spine (Plan, Monitor, Evaluate, Transfer in the research literature) comes from work by Zimmerman, Flavell, and Schraw and Moshman, with practitioner-facing guidance shaped by the Education Endowment Foundation's metacognition reviews and the transfer research of Perkins, Salomon, Bransford, Brown, and Cocking.
Each move the platform offers is mapped to the research that grounds it. The mapping is visible on the teacher dashboard, not hidden in a footer. We treat citing the lineage as the first move, not a footnote.
The four pillars
For students whose thinking outruns their writing, voice is not a feature. It is access. That includes students with dyslexia, students with motor differences, multilingual students, and students with executive function challenges.
Our six-signal model is asset-based by design. It surfaces what students are doing well in their thinking, not what they cannot do on a page. We have committed to a third-party audit of transcription accuracy across speech-pattern variation as part of the V1 process.
The voice pipeline is local. Audio is dropped. No biometric analysis. The platform stores structured evidence, not raw recordings. FERPA and COPPA compliance are baseline expectations, not differentiators.
For disability data specifically, we treat IEP and 504 information with the heightened protection the brief requires. Contracts will reflect data ownership, deletion timelines, and explicit prohibitions on training use.
WCAG 2.1 AA conformance is committed before V1 ships. An Accessibility Conformance Report (VPAT) is on the procurement roadmap. The product offers voice as primary input, with writing, drawing, and assistive input available alongside.
Compatibility with the assistive technology students already use is part of the V1 build, not a later retrofit. Pilot recruitment will prioritize classrooms with strong representation of students with IEPs, students with 504 plans, and multilingual students. The people the product is built for help shape it from the start.
A model card lands with V1 launch and discloses limitations, intended use, training data character, and known bias surface. Our ESSA evidence roadmap targets Tier 3 (promising evidence) before any paid pilot conversion, with a research pathway to Tier 2 over the following 24 months.
Aligned with
- CAST Stewards of Universal Design for Learning since 1984.
- EDSAFE AI Alliance Authors of the SAFE framework: Safety, Accountability, Fairness, Efficacy.
- Educating All Learners Alliance Coalition for inclusive education. Publisher of the April 2026 brief on disability and AI policy.
- New America Education policy research and student data privacy guidance.
- Education Endowment Foundation Independent UK evidence body. Authors of the metacognition and self-regulated learning guidance that shapes how we structure teacher-facing signals.
- Speech Accessibility Project University of Illinois research on speech-pattern variation in AI models.
- Digital Promise UDL Product Certification, in partnership with CAST. Research on equitable edtech.