CogniHuman: A Research Foundation Dedicated to Building Open, Inclusive, and Trustworthy Artificial Intelligence

CogniHuman is a non-profit research foundation registered in India, focusing on advancing research in open-source speech AI and large language models. It is committed to bridging the digital divide, preserving linguistic heritage, and establishing AI ethical standards, so that AI can become a transparent, accessible, and culturally inclusive public good.

The current AI field is dominated by closed-source models from large tech companies, while India has hundreds of languages and dialects. As an Indian Section 8 non-profit company, CogniHuman's mission is to build people's trust in AI by making it more understandable, transparent, and beneficial to everyone.

CogniHuman carries out its work around seven core goals:

1. Advance open research in the intersection of human cognition and AI;
2. Build open-source AI to protect regional language and dialect heritage;
3. Create AI ethical benchmarks and interpretable frameworks that protect human rights and privacy;
4. Provide free/subsidized AI education tools to marginalized communities to alleviate the digital divide;
5. Organize digital skills training to cultivate AI literacy;
6. Maintain a repository of open research results and datasets;
7. Establish a global collaborative ecosystem to address humanitarian challenges.

CogniHuman carries out original research projects:

1. Evaluation framework for Indian language speech AI systems (including standardized test sets, metrics such as WER/MOS/latency, and open-source baseline results);
2. High-quality ethical speech datasets (for underrepresented languages, with complete professional documentation and compliance with consent agreements);
3. Model optimization research (fine-tuning of low-resource language models, CPU-only pipeline optimization, parameter-efficient adaptation techniques like LoRA/QLoRA).

All of CogniHuman's work is open and free: code under Apache 2.0, model weights under Apache 2.0 or MIT, datasets under CC BY-SA 4.0, and documentation under CC BY 4.0. Volunteers are welcome to participate in: data collection, model evaluation, model fine-tuning, documentation improvement, community expansion, etc.

CogniHuman has full recognition from the Indian government: Section 8 non-profit registration, 12A and 80G tax benefits, NITI Aayog (Darpan) registration, CSR-1 eligibility, and INC-20A operational license, providing a solid foundation for cooperation.

CogniHuman benefits from the open-source community (such as Kyutai, Hugging Face, EleutherAI) and represents the possibility of AI technology being open, inclusive, and serving all humanity. True AI progress is not just about increasing model parameters, but also about understanding and respecting every language, community, and culture, providing an example for those concerned with AI ethics, linguistic diversity, and technology for all.