📖 Chapter 7 of 13 · AI for New Learner- Senior 2 · Class 10
Chapter 7: Indian Datasets & Real Problems
In this chapter
- Aryan from Patna explores data.gov.in and AI Kosh.
- Five real Indian datasets you can use this week.
- DPDP Act 2023 — what you legally can and cannot do with data.
- Ethics: bias in Indian datasets. Whose voice is missing?
- Write your first model card. Publish on GitHub.
- Five real Indian AI problems waiting to be solved.
The Big Idea
The best ML models in the world are useless without data. India has unique problems — and the data to solve them is increasingly free, open, and yours to use. The Indian government, IIT labs, and startups have released crores of rupees worth of data. Your job is to find it, use it responsibly, and BUILD something that matters for India.
Did you know?
India’s IndiaAI Mission has allocated Rs 10,372 crore (over USD 1.2 billion) to build Indian AI infrastructure, datasets, and talent by 2027. Half of it is for compute and datasets young engineers like you can access for free.
Builder Profile: Aryan
Patna, Bihar · Aryan, age 14
Aryan studies in Patna. His school doesn’t have an AI lab — but his school has Wi-Fi, and his uncle gave him a 5-year-old laptop. That is all he needed.
He discovered data.gov.in by accident, while looking for Bihar paddy yield data for a geography project. He found 5+ lakh datasets. Free. Public. Crop prices, hospital beds, school enrolment, MNREGA wages — everything. He realised every Indian engineer should know this exists.
He built a Streamlit dashboard showing how flood-affected districts in Bihar got LESS school infrastructure money than non-flooded ones. His teacher submitted it to a state-level science fair. He won. A journalist contacted him. His chart appeared in a news article. He is 14, and his work changed how one Bihar village got funding.
📊 DATASET HUNT Five Indian Datasets to Build With Today
- India Mandi Prices (data.gov.in) — Daily wholesale prices for 350+ crops across 3,000+ mandis. Build: predict tomato prices next month.
- AIIMS-style X-Ray Dataset (openi.nlm.nih.gov) — 11,000+ chest X-rays with disease labels. Build: a TB or pneumonia screener (medical use needs a licence).
- Indian Sign Language — 1,000+ video samples of ISL signs. Build: a real-time ISL-to-text translator, used by 5 crore deaf Indians.
- MIDAS Indic Speech (midas.iiitd.ac.in) — 100+ hours of speech in 8 Indian languages. Build: a voice-command system in Hindi, Tamil, Marathi.
- Atal Bhujal Groundwater (ataljal.mowr.gov.in) — Borehole data from 7 states. Build: predict where wells will dry up, saving farmers and panchayats time.
⚠ FAILURE ANALYSIS
Problem: Indian face-recognition dataset scandal.
Why: In 2022, a US researcher scraped 10 lakh Indian faces from social media without consent. The dataset was used for AI training before being taken down — a massive privacy violation.
Fix: Always ASK before collecting personal data. Document consent. Anonymise where possible. Read DPDP Act Section 4 before any data-collection project.
DPDP Act 2023 — your legal foundation
What the law says, in plain English. Allowed: use anything from data.gov.in for any purpose (it is public and licensed for reuse); train models on aggregated, anonymised data for research if you state your purpose. Not allowed: using personal data (names, Aadhaar, phone numbers, medical records) without explicit consent; scraping Indian websites for personal data, even photos of public figures. Always: honour the right to delete — if a user wants their data removed from your model, you must comply, so build retraining into your workflow.
💻 CODE LAB model_card.md — Template
# Model Card: Mandi Price Predictor ## What this model does Predicts tomato wholesale price in Delhi 7 days ahead. ## Intended users Farmers, traders, agri students. NOT for futures trading. ## Training data data.gov.in Mandi Prices 2019-2024. 2.3M rows. ## Performance RMSE = Rs 4.2/kg. R2 = 0.81 on held-out 2024 data. By region: Delhi=0.85, Mumbai=0.79, Patna=0.71. ## Known limitations - Untested for festival (Diwali, Eid) price spikes - Not validated for crops other than tomato ## Contact & last update Aryan Kumar (Patna) | Updated Mar 2026 ## Licence MIT - free to use, share, modify with attribution.
Five Indian AI problems waiting
Pick one. Build it. Be the engineer who solved it. (1) Crop disease detection in regional languages — most plant-disease apps work only in English, but 40 crore Indian farmers do not; build a Hindi/Tamil/Telugu version. (2) Voice-based UPI for non-readers — 20 crore Indians cannot read but can speak; build voice-only UPI in 10 languages. (3) Wildlife camera-trap classification — forest officers process 50 lakh images a year; build species ID for 100 Indian animals. (4) School dropout early-warning — use anonymised attendance and grades to flag at-risk students before they drop out. (5) Local-language Wikipedia assistant — help Hindi/Marathi/Bengali Wikipedia editors by suggesting edits.
Questions — Chapter 7
- Name 4 real Indian datasets and one thing you could build with each.
- Under the DPDP Act, what may you do freely, and what needs consent?
- Why was the 2022 face-scraping incident a violation?
- What 6 things belong in a model card?
- Pick ONE of the five open problems. Sketch your solution in 3 sentences.
Ship-It Challenge
Pick ONE problem above. Sketch the solution on a single page. Show it to a teacher.
Check your understanding — Chapter 7
Tap an answer to check it instantly. These are for practice — nothing is saved or sent.
Q1. Under the DPDP Act 2023, which is ALLOWED?
- Scraping personal data without consent
- Using open government data from data.gov.in for any purpose — it is public and licensed for reuse
- Selling Aadhaar numbers
Show answer
B. Using open government data from data.gov.in for any purpose — it is public and licensed for reuse
Q2. What are the best free sources of Indian datasets?
- Only paid websites
- data.gov.in, AI Kosh (indiaai.in), Kaggle's Indian section, and the UCI ML Repository
- There are none
Show answer
B. data.gov.in, AI Kosh (indiaai.in), Kaggle's Indian section, and the UCI ML Repository
Q3. What must you ALWAYS honour under the DPDP Act's 'right to delete'?
- Nothing
- If a user wants their data removed from your model, you must comply — build retraining into your workflow
- Only delete data once a year
Show answer
B. If a user wants their data removed from your model, you must comply — build retraining into your workflow
Q4. What must every model card contain?
- Just the accuracy
- Intended use, training data, performance (by sub-group), known limitations, contact, and last update
- Only the author's name
Show answer
B. Intended use, training data, performance (by sub-group), known limitations, contact, and last update