Skip to content

AI Kosh — India's National AI Dataset Repository

CDAC's curated collection of AI-ready datasets in Indian languages and domains — speech, text, images, satellite data.

Provider
CDAC (Centre for Development of Advanced Computing) / MeitY
Website
aikosh.cdac.in
Access Level
🟡 Free Registration Required
Formats Available
CSV, JSON, ZIP (audio/image datasets), API
Coverage
Indian languages, Indian geographies, Indian faces
Use Case
Machine learning, NLP, computer vision, geospatial AI

What Is AI Kosh?

AI Kosh is India's answer to the growing need for Indian-language and India-specific datasets for training AI models. Most global AI models are trained on English/Western data — AI Kosh provides datasets that reflect India's linguistic and geographic diversity.

Available Dataset Categories

Category Examples Size
Speech (Audio) 22 Indian language speech corpora, dialect data TBs
Text (NLP) POS-tagged, NER-tagged Indic text; parallel corpora GBs
Handwriting Devanagari, Tamil, Telugu handwriting recognition GBs
Images Indian face dataset, Indian traffic scenes, Indian food GBs
Satellite Multi-temporal satellite images for land cover classification GBs
Health Medical imaging datasets from Indian hospitals GBs
Documents Scanned Indian government documents, forms GBs

How to Access AI Kosh

Step 1: Register

  1. Go to aikosh.cdac.in
  2. Click "Register" → Fill your name, institution, purpose
  3. Agree to terms of use → Submit

Step 2: Browse and Download

  1. After login: Browse the Dataset Catalogue
  2. Search by: Category / Language / Domain
  3. Click on a dataset → Read the Licence (some are open, some require MOU)
  4. Click "Download" — smaller datasets download directly, larger ones via request

Step 3: API Access (for approved researchers)

AI Kosh provides a REST API for programmatic access:

import requests

# AI Kosh API — requires API token from your profile
API_TOKEN = "your-aikosh-token"
BASE_URL = "https://aikosh.cdac.in/api/v1"

headers = {"Authorization": f"Bearer {API_TOKEN}"}

# List all available datasets
datasets = requests.get(f"{BASE_URL}/datasets", headers=headers)
print(datasets.json())

# Download a specific dataset
dataset_id = "speech_hindi_v2"
download_url = requests.get(
    f"{BASE_URL}/datasets/{dataset_id}/download",
    headers=headers
)

Geospatial AI Datasets on AI Kosh

For readers of this book, these AI Kosh datasets are particularly relevant:

Dataset Description ML Task
ISRO Satellite LULC Labelled satellite images with LULC classes Image classification
Flood Inundation Images Sentinel-1 SAR images + flood labels Semantic segmentation
Indian Road Scene Dashcam images from Indian roads Object detection
Address NER Indian postal addresses with tagged components Named Entity Recognition
Indic OCR Scanned text in 12 Indian scripts Optical Character Recognition

✏️ Practice Exercise

Exercise 6.3 — Find Geospatial Datasets on AI Kosh

Goal: Explore AI Kosh and identify datasets relevant to geospatial analysis.

  1. Go to aikosh.cdac.in and register
  2. Browse the Satellite and Remote Sensing categories
  3. Find at least 3 datasets that would be useful for:
  4. Training a model to classify land cover from satellite imagery
  5. Building an Indian address parser (for geocoding)
  6. Detecting flood extent from SAR images

For each dataset, note: - [ ] Name and description - [ ] Number of samples / size - [ ] Licence terms (open? MOU required?) - [ ] What ML model would you train on this data?

Next: data.gov.in API →