Announcement

M37Labs, an enterprise AI firm with operations in India and San Francisco, announced the launch of Saransh (सारांश), a small language model designed specifically to generate concise, accurate summaries of long‑form Indian news content. Saransh is the first model produced under M37Labs’ Enterprise Proprietary Model (EPM) methodology and is now available for enterprise evaluation.

Technical Details

Saransh is a small language model built from the ground up on M37Labs’ own infrastructure. It was trained from scratch on a corpus of Indian English news articles and reference summaries, using NVIDIA H100 GPUs. The model’s architecture and tokenizer were developed in‑house, and it incorporates native domain vocabulary such as “crore”, “lakh”, and references to Indian institutions like SEBI, MPC and the GST Council. Inference runs on a single GPU and can be deployed on‑premises or within a client‑controlled cloud environment, with zero data egress. Model weights are frozen and versioned, accompanied by a model card for governance.

EPM Thesis and Principles

M37Labs positions Saransh as a demonstration of its EPM thesis, which rests on three core principles: (i) predictable economics – a purpose‑built small model transforms variable, metered inference costs into a predictable infrastructure expense at scale; (ii) data sovereignty – deployment within an enterprise’s own network ensures sensitive documents never leave the organization; and (iii) reproducibility and control – frozen, versioned weights allow consistent outputs over time.

Executive Comments

“Saransh is the first model coming off an AI native firm like M37Labs. We are building narrower, owned, governed models deployable inside the enterprise. That is the M37Labs EPM thesis we are helping enterprises build for their long‑term AI transformation,” said Prashant Shivram Iyer, Co‑Founder and CEO. Zorawar Purohit, Co‑Founder and Chief AI Officer, added, “A model designed for one task has a fundamentally different risk and governance profile from a general‑purpose model… We chose not to fine‑tune our way to Indian news fluency. Saransh learned the language, vocabulary and patterns of its domain from the beginning of training.”

Intended Applications

The model is deliberately limited to abstractive summarization of Indian news and is not intended for general‑purpose chatbot or code‑generation tasks. Potential enterprise uses include media monitoring at scale, leadership and communications intelligence briefs, newsroom and wire support for story digests, and bulk archival processing of historical content where per‑token inference costs would otherwise be prohibitive.

Vertical Roadmap

M37Labs confirmed active development of additional vertical small language models for Retail, BFSI and Healthcare, with releases targeted over the next six months. The Retail model will focus on catalogue, category and consumer‑signal summarization; the BFSI model will address regulatory filings, research notes and advisory correspondence; the Healthcare model will handle clinical documentation and patient‑communication summarization with India‑specific medical terminology.

Availability

Saransh is currently available for enterprise evaluation and can be deployed on‑premises or in a client‑controlled cloud. M37Labs is completing a benchmark against a held‑out evaluation set, which will be included in the model’s technical documentation ahead of broader general availability.

Company Background

Founded in October 2024, M37Labs builds Enterprise Proprietary Models and AI‑powered workflows for global enterprises. Existing vertical AI products include RetailIO.AI for retail intelligence and EBIC.AI for PR and communications intelligence.

Press release distributed by Business Wire India; PTI assumes no editorial responsibility.