Organisation Name

ToumAI

Industry

Financial services, professional services

Organisation Website

https://toum.ai/

Country

Morocco

Sustainable Development Goals (SDGs)

General Description of the AI tool

ToumAI is a multilingual voice-AI platform that enables inclusive, privacy-by-design access to digital and public services. Its lightweight plug-and-play voice assistant works in local dialects and on low-resource devices, giving banks, telcos, transport and public agencies seamless customer interaction. Using pruned and quantised compound AI (5× more efficient than typical LLMs), ToumAI delivers speech-to-text, text-to-speech, emotion and churn analytics across under-represented languages, advancing SDG 11 and related SDGs.

Github, open data repository

https://dvoice.africa/datasets
https://github.com/speechbrain/speechbrain

Relevant Research and Publications

1. 1. Zaouini, M., Chana, L., Frank, R., August, K., Rahmani, Y., Kassis, R., Kumar, S., Pita Costa, J., Orlič, D., Črnko, M., Souss, S., Laleeg, W., Bounouader, Y., Mochariq, O., Elmelsse, Z., Fadil, C., Lamgari, A., Zoubir, M., Doukhou, H., Stopar, L. (2025). Empowering Youth for Smart Cities with AI Solutions to Community and Urban Challenges in the Context of SDG 11. Proceedings of SIKDD 2025, Ljubljana, Slovenia.

Presents ToumAI as a multilingual AI platform bridging the digital divide in Africa and advancing SDG 11; gave early visibility and validation of our approach.

2. Naira, A. M., Benelallam, I., Rahmani, Y., Bahafid, A., & Erraji, Z. (2023). Data Creation and Empirical Evaluation of Cross-Lingual Learning in Comorian Dialects. NeurIPS 2023 Workshop on African Natural Language Processing.

Built and open-sourced datasets for low-resource African dialects, validating our transfer-learning approach to under-represented languages and influencing ToumAI’s data pipeline.

3. Naira, A. M., Benelallam, I., Erraji, Z., & Bahafid, A. (2023). BantuLM: Enhancing Cross-Lingual NLP in the Bantu Language Family. In Proceedings of the International Conference on African Language Technologies.

Developed a pre-trained BERT model for 18 Bantu languages, showing that training on related languages improves performance — the basis of ToumAI’s compound AI architecture.

4. Naira, A. M., Benelallam, I., Rahmani, Y., Bahafid, A., & Erraji, Z. (2024). Geopolitical News Assessment on African Countries Using Multilingual LLM and Public Attention Methodologies. Elsevier publication.

Demonstrated domain-adapted multilingual models on African languages, informing ToumAI’s analytics and ethical AI practices for public-sector use cases.

Needs