AIKosh
Why in News?
IndiaAI Mission held an AIKosh workshop on 20 August, 2026 to strengthen India’s AI‑ready data ecosystem, and ran AIKosh University Engagement Programmes to train students and encourage dataset contributions.
Purpose & Ecosystem
- National AI Repository: AIKosh serves as India’s central unified data platform under the Ministry of Electronics and Information Technology (MeitY).
- Current Scale: The repository hosts over 15,000 datasets, 300 AI models, and 30 toolkits.
- Target Users: It directly supports Indian researchers, startups, developers, and students.
- Sovereign AI Goal: It focuses on building domestic capabilities to reduce reliance on foreign AI systems.
Features & Technical Offerings
- Shared Compute Power: Integrates with the IndiaAI Compute Platform to offer cloud-based GPU resources, including NVIDIA A100 allocations.
- Secure Sandbox Environment: Developers can build, train, and test models seamlessly using an integrated development environment.
- Cross-Sector Consolidation: Aggregates data across 20+ sectors like healthcare, education, and agriculture.
- Robust Security Infrastructure: Protects assets using content discoverability, permission-based access, data encryption, and real-time firewalls.
Focus on Indic Languages & Foundation Models
- Sovereign Models: Hosts indigenous foundational models like Sarvam 30B/105B, BharatGen Param2, and Vachana.
- Massive Datasets: Features IndicVoices (12,000 hours of multilingual speech across 22 languages).
- Speech Translation: Integrates BhasaAnuvaad, the largest speech translation dataset for Indian languages.
Safe & Trusted AI Initiative
- Ethical AI Tools: Set to deploy four specialized tools for machine learning, bias mitigation, risk evaluation, and fairness assessment.
- Open-Source Roadmap: Future tools for deepfake detection and AI content watermarking will be released as open-source assets.
- Traditional Medicine Synergy: Partnered via an MoU with the Ministry of AYUSH to digitize and securely onboard anonymized traditional healthcare datasets.
