Indian startups turn to small language models to solve for efficiency, privacy, cost

Indian startups are increasingly prioritizing small language models (SLMs) over large language models (LLMs) to navigate unique challenges within the Indian market. These challenges include high cloud costs, limited digital infrastructure, and stringent data privacy regulations like the Digital Personal Data Protection (DPDP) law. While LLMs excel at broad tasks, SLMs—typically ranging from 1 to 15 billion parameters—are more effective for sector-specific applications requiring accuracy and data security.

Several companies, including Stockgro, Dhan (with its Artham model), August, Qure.ai, Powerplay, and Gnani AI, are developing and deploying custom SLMs. These models are often trained on proprietary data and hosted on-premise or locally to maintain data privacy and reduce reliance on cloud services. This approach allows for real-time analysis, even in areas with limited or no internet connectivity, as demonstrated by Qure.ai’s use of edge models in rural healthcare.

Experts like Sourav Banerjee of Shunya Labs emphasize that SLMs are more cost-effective, faster, and secure for specialized tasks compared to LLMs. The shift towards SLMs reflects a growing recognition that smaller, targeted models can deliver significant value while addressing the specific needs and constraints of the Indian startup ecosystem.

(Source：The Economic Times)