AI Masterclass in Public Services

Abstract

The masterclass aimed to demonstrate how public‑sector AI solutions can be conceived, designed and prototyped in under an hour without writing a single line of code. Arun and Sankha opened with a vision of AI as a civilizational and ethical shift for India, outlining three guiding principles—AI‑first systems, filling the “missing middle” for frontline workers, and building on a sovereign technology stack. They then prescribed four design choices (vernacular voice‑first interfaces, conversational rather than form‑based interactions, proactive outreach, and frugal‑yet‑sovereign deployment). The remainder of the session was a live, participant‑driven lab in which attendees used a natural‑language prompting platform to model problems (e.g., ASHA‑worker data capture), iteratively refine a product‑requirements document (PRD), and generate a functional prototype. Multiple participant demos (document‑verification, nutrition‑analytics, snake‑bite guidance) illustrated the breadth of possible public‑service applications. The session closed with three strategic takeaways for scaling AI‑first public services in India.

Detailed Summary

Arun Ramchandran opened the masterclass with a formal welcome to dignitaries, senior government leaders, TCS colleagues and participants. He positioned AI as a transformative juncture for nation‑building and governance, likening it to past “revolutions” (industrial, digital) and emphasizing that the choices India makes now will set a global tone.

Key points articulated by Arun:

• AI is more than a technology shift; it is a civilizational change that reshapes ethics, morality, and societal structures.

• While AI will be built worldwide, India must ensure the AI built here is ethical, uplifting and “Indian‑minded.”

• Three guiding principles were introduced:

  1. AI‑first systems – moving beyond purely digital platforms that wait for citizens to “knock.”
  2. The “missing middle” – empowering frontline workers (e.g., ASHA, Anganwadi, postal staff) with intelligent assistants that fuse human empathy with expert knowledge.
  3. Sovereign stack – avoiding a “borrowed brain” by owning data, models, infrastructure and tooling.

Arun underscored India’s current digital metrics: 38 % digital literacy, 38 000 GPUs allocated to the national AI mission, and the emergence of INDIQ vernacular models that already match or surpass global benchmarks.

2. Expanding the Guiding Principles

2.1 AI‑First Systems

• Current digital services are reactive: they require citizens to fill forms, accept limited language inputs, and often rely on intermediaries.
• AI‑first means services anticipate needs, reach out proactively, and converse in the citizen’s language and context.
• An illustrative vignette: a system that senses a “deficit monsoon” and proactively contacts vulnerable farmers, rather than waiting for a complaint.

2.2 The Missing Middle

• India’s knowledge concentration remains urban, while 80 %+ of the population lives rural.
• Frontline workers (≈ 1 million ASHA workers) possess empathy but lack digital tools.
• AI can become a bridge: it digitises shared context, aggregates specialist knowledge, and delivers vernacular, frugal‑device‑compatible assistants.
• Crucially, AI will not replace these workers; it will augment them, pairing human empathy with algorithmic intelligence.

2.3 Sovereign Stack

• The AI stack comprises infrastructure → data → models → platforms → applications.

• To retain sovereignty, India must:

  1. Host training data on Indian data‑centres.
  2. Use open‑weight models (even non‑Indian models) so they can be inspected and run locally.
  3. Prefer open‑source or India‑first tooling.

• Initiatives cited: AI‑Kosh (large, curated Indian datasets), INDIQ language models, and the government’s GPU allocation for Indian innovators.

3. Prescriptive Design Choices for Public‑Sector AI

Arun distilled the discussion into four concrete design decisions that participants were asked to embed in their prototypes:

#	Design Choice	Rationale
1	Voice‑first & vernacular interfaces – speech‑driven, capable of understanding 22 Indian languages and dialects.	Overcomes low digital literacy and language barriers.
2	Conversational flows, not static forms – the system asks questions, gathers context dynamically.	Reduces friction, eliminates repetitive data entry.
3	Proactive outreach – systems initiate contact, provide guidance, and anticipate needs.	Shifts from reactive to citizen‑centric service.
4	Frugal & sovereign deployment – solutions must run on commodity hardware while staying on Indian data‑centres.	Ensures scalability across resource‑constrained environments.

4. Transition to the No‑Code Lab

Arun announced the “wipe‑coding” session: participants would apply the four design choices to a curated challenge, building a functional AI prototype using natural‑language prompts (English, with vernacular extensions).

• Goal: compress the ideation‑to‑prototype cycle from months to ≈ 40 minutes.
• Method: a prompt‑driven platform (later identified as Sarvam AI on the Lovable rapid‑build environment).
• Facilitators: Arun (lead), Sankha Som (co‑facilitator), and a team of on‑site volunteers.

4.1 Audience Warm‑Up

Sankha asked the room: “How many of you have ever written code?” The response was far higher than expected, signalling a comfortable baseline for participants to engage with the AI‑prompting workflow.

4.2 Core Principles for Prompt Engineering

• Emphasised the “artificial” part of AI: the model can accelerate but human guidance remains essential.
• Demonstrated prompt‑failure (hallucination) with a flawed email‑drafting example, illustrating why iterative clarification is mandatory.
• Highlighted the computational leap from 1975’s 5 MHz satellite processors to today’s smartphones (≈ 1 000× more compute), explaining why AI is now accessible.

4.3 Human‑Centred Prompting Workflow

Arun walked through the four‑phase workflow (mirroring the earlier guiding principles):

1. Understand – AI asks participants to identify who is involved, what the pain points are, and why they exist.
2. Diagnose – AI hypothesises root causes, prompting participants to accept, reject, or amend each hypothesis.
3. Imagine – AI proposes a solution concept (e.g., a voice‑driven data‑capture app for ASHA workers). Participants refine the proposed flow.
4. PRD Generation – AI compiles a Product Requirements Document (problem statement, solution overview, user personas, key features, user‑flow) in plain English.

The resulting PRD can be fed back into the platform to auto‑generate a prototype (UI mock‑up, basic workflow, voice‑recognition backend) within minutes.

5. Live Demonstration – The ASHA Worker Use‑Case

5.1 Problem Definition

• ASHA workers (≈ 1 M) currently record health data on paper, then later manually enter it into a government portal, causing double work and data latency.
• Desired outcome: single‑step, vernacular voice entry that updates the portal in real time, increasing daily coverage from 6–7 households to a much higher number.

5.2 Prompt‑Driven Exploration

• The AI listed stakeholders (ASHA, auxiliary nurses, health ministry, rural families) and asked participants to confirm or edit the list.
• It then queried each stakeholder’s pain points (e.g., “ASHA double entry,” “access to real‑time data,” “technical reliability”). Participants added missing concerns (e.g., “network connectivity”).
• Once the “who” and “why” were captured, the AI moved to the diagnose stage, proposing root causes (paper‑based workflow, lack of vernacular interfaces).

5.3 Imagining the Solution

• AI suggested a voice‑driven mobile app that:

  • Accepts speech in the worker’s native language (via Sarvam AI’s 22‑language engine).
  • Transcribes and stores data instantly on a sovereign cloud.
  • Generates real‑time dashboards for health officials.

• Participants iteratively refined the flow (adding offline caching, error‑checking).

5.4 PRD Generation & Rapid Build

• The AI produced a PRD that listed:

  • Problem statement – manual double entry for ASHA workers.
  • Solution overview – voice‑first, vernacular app on frugal devices.
  • Target users – ASHA workers, health officials, district administrators.
  • Key features – speech‑to‑text, multilingual UI, offline sync, analytics dashboard.
  • User flow – open app → select language → dictate patient data → auto‑populate fields → submit → confirmation.

• The PRD was downloaded and fed into Lovable, which immediately generated a working prototype (screen mock‑ups, voice capture UI, basic backend).

• Arun highlighted that no code was required; the entire pipeline was natural‑language driven.

6. Participant Prototypes – Showcase

After the ASHA walk‑through, volunteers from several tables presented the prototypes they had built.

6.1 Document‑Doctor (GIZ/World Bank)

• Use‑case: Verify citizen‑submitted documents (birth certificate, utility bill) before a service appointment.
• Functionality: AI flagged mismatched names, outdated documents, and suggested missing address proof.
• Outcome: Demonstrated that AI can pre‑screen documents, reducing errors at the point of service.

6.2 Nutrition Analytics for Schools (Goa Higher Education)

• Problem: Uniform mid‑day meals across districts ignore regional nutrition needs.
• Solution: Teachers use a voice‑driven app to record student height, weight, BMI; AI aggregates data, identifies under‑nutrition hotspots, and informs kitchen staff and government.
• Key Insight: Dynamic, vernacular data capture enables real‑time nutrition policy adjustments.

6.3 Snake‑Bite Guidance App

• Problem: Rural health workers need instant, language‑specific SOPs for snake‑bite emergencies.
• Solution: Voice‑activated app accepts parameters (child age, weight, snake type) and returns step‑by‑step treatment in the worker’s language.

6.4 Additional Quick Wins

• A “Document Doctor” that detects errors in uploaded PDFs (e.g., a Canadian birth certificate flagged as mismatched).
• An assessment‑order helper for tax‑related queries, showing how AI can read legal documents aloud and allow on‑the‑fly corrections.

Across all demos, participants repeatedly emphasized the ease of iterating: they could change a single word in the PRD, re‑run the generator, and instantly see the UI update.

7. Closing Remarks & Call to Action

Arun returned to the stage for the final minutes:

Key Takeaways

• Declared that AI is now a workplace skill, comparable to PowerPoint or Excel.
• Invited participants to stay connected with the TCS AI‑public‑services team, offering email contacts for ongoing support.

A brief applause marked the end of the formal session; the room was asked to stay for a few minutes while volunteers completed their demonstrations.

---

---

See Also:

• democratizing-ai-resources-and-building-inclusive-ai-solutions-for-india
• ai-for-inclusive-societal-development
• flipping-the-script-how-the-global-majority-can-recode-the-ai-economy
• consumers-at-the-core-building-ai-with-consumer-centric-approach