What India's 190 AI projects still need - BusinessToday

Global AI models often fail in India because they rely on limited, off-the-shelf datasets that ignore local linguistic diversity and physical realities. These generalist recordings miss crucial nuances, such as code-mixed speech in support calls, handwritten regional scripts, or specific retail packaging. Consequently, deployed solutions struggle to function in everyday contexts, leading to operational failures, increased human escalation costs, and pilots that never transition from demonstration to real-world utility. The solution lies in commissioning bespoke, on-request data collection rather than relying on generic archives. Companies must provide specific briefs detailing the required languages, districts, and data types, ensuring that recordings are made by local contributors in the actual environments where the AI will operate. This approach generates high-fidelity, context-aware data across various modalities, including audio, video, and sensor inputs, directly addressing the gaps that standard global datasets cannot fill. This distinction is vital for open data initiatives because it highlights that open datasets must be granular, representative, and locally grounded to be truly useful. The relevance lies in moving beyond broad, aggregated data to specific, permission-based collections that reflect the complexity of real-world applications. For the open data community, this underscores the need for infrastructure that supports diverse, localized contributions, ensuring that AI systems are inclusive and effective across all demographics, not just those in well-documented regions.

Source: businesstoday.in
Published on 2026-10-22

What India's 190 AI projects still need - BusinessToday

Related news