Public Systems & Education ASR Speech Tech Programme Diagnostics ASER 2022 Grounded

FLN Check: A Digital Foundational Learning Diagnostic & Intervention Engine

An independent public policy project demonstrating how adaptive assessment design, speech recognition technology (Sarvam AI), and cohort data pipelines can accurately diagnose foundational learning gaps and immediately turn them into targeted classroom interventions.

Try Interactive Diagnostic Explore Live Community Stats

1. The Public Policy Challenge

Across India, the National Education Policy (NEP 2020) and the NIPUN Bharat Mission have identified Foundational Literacy and Numeracy (FLN) as an urgent national priority. However, conventional testing architectures suffer from three critical systemic failures:

Binary Examination Mindset
Standard exams reduce children to percentages (e.g. 48/100) rather than identifying the precise conceptual ceiling a child can comfortably perform.
Diagnostic–Intervention Disconnect
Assessment data is collected months after instruction and rarely loops back to the child or teacher with actionable micro-practice.
Reading Oral Evaluation Bottleneck
Evaluating reading requires 1-on-1 human listening. In classrooms of 40+ students, oral assessments are rarely conducted with frequency.

2. Grounded in the ASER Assessment Methodology

FLN Check is informed by the proven, gold-standard ASER (Annual Status of Education Report) field assessment methodology. Rather than creating arbitrary questions, the tool codifies the progressive difficulty ladders from the official ASER 2022 instruction manual:

Reading Hierarchy (Progressive)

Beginner Level
↓ Letter Recognition (≥ 4 / 5 correct)
↓ Word Reading (≥ 4 / 5 correct)
↓ Paragraph Reading (≤ 3 errors, sentence cadence)
↓ Story Reading (≤ 3 errors, fluent & ease)

Arithmetic Hierarchy (Progressive)

Beginner Level
↓ Number Recognition 1–9 (≥ 4 / 5 correct)
↓ Number Recognition 11–99 (≥ 4 / 5 correct)
↓ 2-Digit Subtraction with Borrowing (2 / 2 correct)
↓ 3-Digit by 1-Digit Division (Quotient & Remainder)

3. Speech Technology & ASR Scoring Pipeline

One of the major technical pillars of this project is solving speech recognition in low-resource Indian languages and accents without penalizing children for acoustic artifacts or local dialect variants.

Speech-to-Diagnosis Data Pipeline
1. Browser Audio
MediaRecorder 16kHz WAV/WebM
2. Sarvam AI ASR
Model `saaras:v4` Indic STT
3. Normalization
Punctuation & script canonicalization
4. ASER Alignment
Dialect & synonym tolerance rules
5. Profile Output
Highest demonstrated level
Critical Technical Principle: Speech recognition confidence is strictly separated from reading accuracy. If background noise or low acoustic confidence occurs, the system gently prompts for a retry rather than falsely flagging the child with reading deficits.

4. Relevance to Public Systems Leadership

For a Founder’s Office or social-sector programme-management mandate, this project illustrates an end-to-end execution loop:

Systems Thinking: Translates national FLN policies into empirical, field-tested measurement criteria.
Data Privacy by Design: Collects zero child PII (no names, phone numbers, or Aadhaar), using anonymous avatar identifiers.
Intervention Slicing: Aggregates classroom cohorts into actionable priority cards (e.g. 13 students requiring borrowing practice).
Longitudinal Reassessment: Evaluates baseline-to-endline skill gains to assess actual instructional delta.