MTSS Reading Platform

Standard speech recognition systems are trained on adult voices. They fail with young children, who have different vocal patterns, pronunciation inconsistencies, and background noise from classroom environments.

We built a custom phoneme recognition pipeline on Microsoft Azure Cognitive Services, training it specifically on children's speech from the pilot group. The system needed to distinguish between correct and incorrect phoneme responses, not just transcribe words.

A child reading the word "cat" might say "cuh-ah-tuh" or "kat" or "cot." The system needs to identify whether they correctly pronounced the /k/ sound, the short /a/, and the /t/. This is different from transcription, which would simply convert speech to text.

We trained the model to detect correct vs. incorrect phoneme responses with 85% accuracy. This level of accuracy is sufficient for formative assessment, where the goal is to identify patterns and guide instruction, not to replace human judgment entirely.

Simple phoneme detection runs on-device using TensorFlow Lite. When a kindergartener taps a letter and says the sound, the system provides instant feedback without network latency.

Complex sentence reading and fluency assessments run in the cloud. These tasks require more processing power to analyze pacing, intonation, and word accuracy. The hybrid approach balances responsiveness with capability.

The platform includes an adaptive assessment engine that adjusts difficulty in real time based on student performance. If a child struggles with letter sounds, the system doesn't advance to blending. If they master phoneme recognition quickly, it moves to more complex tasks.

This prevents frustration and disengagement. A student who cannot yet recognize the letter "B" will not be asked to read words with "B" in them. The system meets each child at their current skill level.

The assessment and instruction align with the Big Five components of reading: phonemic awareness, phonics, fluency, vocabulary, and comprehension. This structure matches Science of Reading principles and MTSS frameworks that schools already use.

Teachers see real-time dashboards showing each student's progress across these five areas. They can identify which students need intervention in phonemic awareness vs. fluency, and group students accordingly.

The interface is designed for pre-literate children. Large tappable buttons, audio-driven instructions from an avatar, and visual cues allow PreK and kindergarten students to use the system independently after initial training.

This amplifies teacher capacity. Instead of spending 20 minutes assessing one child, a teacher can monitor a classroom of students working through adaptive exercises, intervening only when the system flags a student who needs help.

Alpha had thousands of pages of evidence-based print curriculum materials, including Reading Mastery and Acadience. These materials are pedagogically sound but not digital.

We built an OCR and automated content processing pipeline that converts PDFs and images into interactive JSON modules. The system maintains pedagogical fidelity, preserving the instructional sequence and scaffolding from the original materials.

The pipeline is designed for ongoing content ingestion. As Alpha adds new curriculum materials or updates existing ones, the system processes them automatically.

This flexibility allows the platform to support multiple curricula and adapt to different state standards. A district using a specific phonics program can integrate that content without custom development work.

The result is a content engine that transforms static print materials into adaptive digital experiences, complete with speech recognition, immediate feedback, and progress tracking.

The platform handles sensitive student data, including audio recordings of children's voices and individual performance metrics. We designed the system with privacy as a core requirement, not an afterthought.

Audio recordings are processed and then discarded. Only the assessment results are stored. Student data is encrypted in transit and at rest, with access controls that limit who can view individual student information.

This approach meets FERPA requirements and addresses parent concerns about student privacy. Schools can adopt the platform without compromising their data governance policies.