⚡ Lightning Talk · CPP Conference 2026

Trustworthy AI in Education:
Human Oversight Lessons from India

A 15-minute synthesis of two longitudinal case studies — from a restaurant chain in Bengaluru to a school for the blind in Chikkamagaluru — with evidence-based policy recommendations for Indian government EdTech.

Author Dr. Smrite Goudhaman
Conference CPP Conference 2026
Format Lightning Talk · 15 min
Affiliation Golden Gate University · Datamatics Global Services

When Classrooms Become the Lab

This lightning talk bridges two independent longitudinal case studies — one in enterprise learning, one in disability education — to surface a shared finding: human oversight is not a compliance requirement. It is the design condition under which AI in education actually works.

Study one tracked AI-enabled micro-agent deployment at Toscano Restaurant Chain (Bengaluru), where a cohort of 100 employees scaled to 250 over two years. Study two documented a two-week AI literacy pilot at Asha Kirana School for the Blind (Chikkamagaluru, Karnataka) — a school established in 1990 that has never had access to tools designed for its students.

Both studies produced the same underlying lesson: AI systems that embed human oversight into their architecture — rather than treating oversight as an audit layer bolted on after deployment — achieve measurably better outcomes for learners. The talk ends with three policy recommendations for India's national EdTech strategy.

Trustworthy AI Human Oversight AI in Education India EdTech Inclusive Learning Disability Education Constrained Agency

Two Studies. One Finding.

Case Study 1 · Enterprise Learning
Toscano Restaurant Chain — Bengaluru

AI-enabled micro-agent deployment via the SafetyCulture platform. Longitudinal cohort tracked 2024–2025 across two scaling phases.

86.82%
Course completion rate
+4.26
Assessment score improvement (pts)
−57%
Time-on-task reduction
250
Employees by Year 2

Key concept: Constrained Agency — the principle that AI micro-agents operating within defined human-supervised boundaries outperform unconstrained systems on both adoption and outcome metrics.

Case Study 2 · Disability Education
Asha Kirana School for the Blind — Chikkamagaluru

National Ekalvya Award for PwD Education. Est. 1990. Two-week AI literacy intensive pilot (Goudhaman & Srivastava, 2026) with Grade 10 graduates and Classes 2–9.

49
Students enrolled
7/9
Grade 10s using ChatGPT independently by Module 2
2
Learning tracks (Voice-First & Visual-First)
100%
Offline-capable curriculum

Key finding: Students who had never interacted with AI tools achieved independent use within days — when curriculum was designed for their modality, not adapted from sighted content.

"For the first time, they felt at par with the technological advancements of their sighted peers."

— Dr. Varsha, Principal, Asha Kirana School for the Blind · Chikkamagaluru, Karnataka

The Oversight Architecture Hypothesis

Both case studies point to the same structural finding: AI deployments that treat human oversight as an architectural feature — embedded in how the system works, not layered on after — produce better learning outcomes than systems where oversight is an afterthought.

In the Toscano deployment, human managers remained visible in the feedback loop throughout the micro-agent interactions. In the Asha Kirana pilot, instructors co-designed the curriculum modality with the students. In both cases, human presence was not just an ethical safeguard. It was the mechanism of learning.

Principle 1
Trust Before Capability
Learners must trust the system before they will engage with it. Capability demonstrations alone do not generate adoption. Transparency about how the AI works — and what it cannot do — increases trust faster than performance metrics.
Principle 2
Modality-First Design
Accessible AI education cannot be sighted-first content with accessibility overlays. Voice-first and visual-first tracks must be designed natively for their learners, with human educators co-designing the curriculum alongside students with lived experience.
Principle 3
Constrained Agency as Pedagogy
AI agents that operate within defined boundaries — with human oversight visible in the interaction loop — outperform unconstrained systems on both completion rates and assessment outcomes. Constraint is not a limitation. It is a design feature.

Three Recommendations for India's EdTech Strategy

Recommendation 1
PMKVY AI Qualification Pack for PwD
Develop a dedicated AI Qualification Pack within the Pradhan Mantri Kaushal Vikas Yojana framework specifically for Persons With Disabilities — with modality-native curriculum design, offline-capable delivery, and human mentor integration as a mandatory component, not an add-on.
Recommendation 2
WCAG 2.1 AA Mandate for Govt EdTech
All government-funded EdTech platforms must meet WCAG 2.1 Level AA accessibility standards as a condition of procurement. Current platforms were not designed for students like those at Asha Kirana — and the gap in outcomes is directly traceable to this design failure.
Recommendation 3
Human Oversight as Deployment Criterion
AI deployments in publicly funded educational institutions should be evaluated on the presence and quality of human oversight architecture — not solely on capability benchmarks. The Toscano and Asha Kirana data both support this as a leading predictor of learning outcomes.

What the Two Studies Share

Dimension Toscano Restaurant Chain Asha Kirana School
ContextEnterprise vocational learning (F&B)Disability education (Class 2–10)
AI SystemSafetyCulture micro-agent platformCustom modality-native curriculum (ChatGPT + Web Speech API)
Oversight ModelManager-visible feedback loopTeacher co-design + real-time facilitation
Cohort100 → 250 employees (2 years)49 students (2-week intensive)
Key Outcome86.82% completion, −57% time-on-task7/9 Grade 10s independent ChatGPT use by Module 2
Shared FindingEmbedded human oversight → measurably better learning outcomes

Dr. Smrite Goudhaman

Doctor of Business Administration, Golden Gate University, San Francisco (2025). Research focus: AI governance, human oversight frameworks, and enterprise AI adoption. Published across MERC 2026 (IIM Kashipur), CERE 2026 (IIM Indore), and CPP Conference 2026.

Pre-Sales Solutions Leader — AI & Enterprise Transformation, Datamatics Global Services. Adjunct Faculty, Ageno School of Business, GGU. Faculty Member, upGrad.

Speaking record includes: UN AI for Good Global Summit (Geneva, July 2026), IASEAI at UNESCO Headquarters (Paris, February 2026), IIT Delhi AI Impact Summit, AAAI-26 (Singapore), Pearl Academy, Kristu Jayanti University, Christ University.

Recognition
Awards & Distinctions
  • ⭐ Dr APJ Abdul Kalam Inspiration Award 2026
  • ⭐ Naari Shakti Award 2026
  • ⭐ South India Women Achievers Award 2026
  • ⭐ Global 200 Women Power Leaders 2026 — White Page International
  • ⭐ Hindustan Book of World Records — Women's Empowerment
  • ⭐ Oxford Saïd Business School & UNESCO — AI in Government (Certified)

Cite & Connect

Primary Citation

Goudhaman, S. (2026). Trustworthy AI in Education: Human Oversight Lessons from India. CPP Conference 2026, Lightning Talk.

Goudhaman, S., Dahiya, T., & Kumar, S. (2026). IT Human Oversight Frameworks for Agentic AI. MERC 2026, IIM Kashipur. Paper ID MERC2026-0093 IT.

Goudhaman, S., Dahiya, T., & Kumar, S. (2026). From Pilot to Scale: AI-Enabled Micro-Agents. CERE 2026, IIM Indore. Paper 41.

Goudhaman, S. & Srivastava, R. (2026). AI for Every Mind: Designing Accessible AI Curricula for PwD. CPP Conference 2026, IIM Bangalore.

UNESCO (2021). Recommendation on the Ethics of Artificial Intelligence.

Ministry of Education, India (2023). National Digital Education Architecture (NDEAR).

Web Content Accessibility Guidelines (WCAG) 2.1. W3C Recommendation, June 2018.

PMKVY 4.0 (2022–2026). Pradhan Mantri Kaushal Vikas Yojana Framework. Ministry of Skill Development.

Edelman (2024). Trust Barometer: AI and the Future of Trust.

Floridi et al. (2018). AI4People — An Ethical Framework. Minds and Machines, 28(4), 689–707.

NIOS (2023). Guidelines for Accessible Education for Persons With Disabilities. National Institute of Open Schooling.