National AI Initiative

Advancing Indigenous AI Foundation Models

ENTTROPIA is an open-science AI research foundation developing multilingual models, computing tools, and policy frameworks for public-interest transformation in India.

Section 8 Not-for-Profit

Open Access Datasets

IndiaAI Mission Aligned

Explore Projects Publications

// Foundation Manifesto

Open research for public-interest intelligence.

We develop indigenous models, accountable AI systems, and shared scientific infrastructure for India—independent of commercial profit motives.

Foundation, pillars & mandate

// Current Research

Foundation Models / NLP

ACTIVE

Indic Multilingual Foundation Model (IMFM-7B)

Development of a 7-billion parameter multilingual large language model pre-trained on 22 scheduled Indian languages including Hindi, Bengali, Tamil, Telugu, Kannada, and Odia. The model uses a custom BPE tokenizer optimised for Devanagari and Dravidian script families, targeting instruction-following and reasoning tasks across low-resource Indian language pairs.

LLMMultilingualIndicFoundation ModelsNLP

Lead: Team Member 01, Team Member 04Since 2025

Responsible AI / Policy

ACTIVE

Responsible AI Audit Framework for Public Sector Deployments

Design and validation of a systematic audit framework for assessing fairness, accountability, transparency and explainability (FATE) in AI systems deployed in government and public-interest contexts. Includes automated bias detection toolchains, stakeholder impact assessments, and regulatory compliance checklists aligned with the Digital India Act.

Responsible AIFairnessPolicyAuditGovernance

Lead: Team Member 02, Team Member 05Since 2025

Full research index

// Recent Publications

arXiv preprint

[2025]

IMFM-7B: A Multilingual Foundation Model for 22 Scheduled Indian Languages

Team Member 01 · Team Member 04 · Team Member 05

We present IMFM-7B, a 7-billion parameter causal language model pre-trained on a 1.4T token corpus spanning 22 scheduled Indian languages. We describe the tokenizer design, data curation pipeline, and evaluation across IndicGLUE and Flores-200 benchmarks. IMFM-7B achieves state-of-the-art zero-shot performance on 14 of 22 target languages.

LLMMultilingualIndicPretraining

DOI: 10.48550/arXiv.2025.00001

ACM FAccT 2025

[2025]

A FATE Audit Framework for Government AI: Design, Implementation, and Case Studies

Team Member 02 · Team Member 05 · Team Member 07

This paper presents a comprehensive audit methodology for evaluating Fairness, Accountability, Transparency, and Explainability in AI systems used by the Indian public sector. We operationalise each FATE dimension as measurable metrics and apply our framework to three real-world deployments, revealing systemic biases and recommending corrective actions.

Responsible AIFairnessAccountabilityAudit

DOI: 10.1145/facct.2025.00123

All publications

// Latest Insights

Indic language streams converge into a luminous neural map of India

12 Oct 2025 3 min read

Why Indic LLMs Matter: The Case for Indigenous Language AI

By Team Member 01

Over 1.4 billion people speak one of India's scheduled languages, yet less than 2% of large language model training data represents them. This essay makes the case for urgent, systematic investment in Indic foundation models.

LLMIndic LanguagesAI Policy

Four transparent audit lenses guide a civic decision system

28 Nov 2025 5 min read

Responsible AI for Government: What FATE Really Means in Practice

By Team Member 02

Fairness, Accountability, Transparency, and Explainability sound like abstract principles. This post describes what they mean concretely when applied to an AI system deciding citizen benefit eligibility.

Responsible AIPolicyGovernanceFATE

Read archive