MirasAI

About Us

MirasAI LLC is an NLP research and language data company based in Bloomington, Indiana, founded and led by Meesum Alam, a Fulbright PhD candidate in Computational Linguistics at Indiana University Bloomington. MirasAI is built around a simple premise: the languages of South Asia — spoken by well over a billion people — remain drastically underrepresented in the datasets that power modern language technology. We exist to close that gap.

The company operates two divisions. Our Research & Data division builds and publishes open speech, text, and multimodal datasets covering 60+ South Asian languages, including Urdu, Punjabi, Saraiki, Sindhi, Pashto, Balochi, Burushaski, and others. Our Technology Services division applies this expertise commercially, delivering language data and voice AI solutions to enterprises and developers working in the region.

Our team combines academic rigor with on-the-ground fieldwork experience across Pakistan and the wider South Asian linguistic landscape, including work with speaker communities, universities, and local partners to source, annotate, and validate data responsibly.

Our Mission

MirasAI's mission is to make South Asia's linguistic diversity visible and usable in AI — not as an afterthought, but as a first-class part of how language technology is built. Hundreds of millions of speakers of Saraiki, Sindhi, Pashto, Balochi, Burushaski, and other low-resource languages currently have little to no representation in the datasets that train speech recognition, translation, and language models. We believe that gap is both a technical failure and an equity failure, and we're working to close it through rigorous, ethically sourced, and openly available data.

Concretely, this means:

• Building foundational datasets for languages that have never had structured digital resources before
• Publishing peer-reviewed research alongside our datasets (e.g., LREC, MWE-UD, SIGUL) so the data meets academic standards of quality and reproducibility
• Prioritizing consent, fair compensation, and community involvement in data collection
• Making datasets genuinely usable — well-documented, properly licensed, and accompanied by clear datasheets

Why We Share Data on Mozilla Data Collective

We share our datasets on Mozilla Data Collective because MDC's mission — open, ethically sourced data for underrepresented languages — is directly aligned with our own. Publishing through MDC lets us:

• Reach the research and developer communities most likely to use and build on low-resource South Asian language data
• Benefit from MDC's infrastructure and standards for hosting, licensing, and documenting large-scale datasets
• Contribute to a growing, trusted body of open language resources rather than keeping this work siloed within a single company or product
• Support the broader goal of linguistic equity in AI — ensuring that speakers of Saraiki, Pashto, Balochi, Burushaski, and other South Asian languages are not left behind as language technology advances

To date, our contributions to and work through MDC include large-scale speech and text corpora (Punjabi, Saraiki, Shina), the Burushaski-English Speech Translation Corpus, the Associated Press Pakistan multilingual news corpus, and translation of Meta's BOUQuET benchmark into seven Pakistani languages. We see MDC not just as a hosting platform, but as a partner in the long-term project of building a truly multilingual data commons for South Asia.