Natural language processing

Text-to-speech (TTS) – Qfrency

Qfrency is our flagship Text-to-speech software that converts digital text into natural-sounding speech. It is designed for cloud-based and offline use, supporting a wide range of applications, from assistive technologies (like screen readers, navigation aids and, book readers) to customer service interfaces and call centres.

NLP literacy enablement – Ngiyaqonda

Ngiyaqonda harnesses NLP to enhance reading and writing skills in the official languages of South Africa. It offers personalised learning paths and interactive exercises for proficiency at scale. Our focus with Ngiyaqonda is on creating inclusive technologies that break language barriers and promote digital access.

Controlled Natural Language Processing

We develop modern computational grammar for the South African languages. Our grammars enables syntactic parsing that has otherwise not been achieved for these languages, as well as domain specific language generation.

Automatic Speech Recognition

Our automatic speech recognition solution offers highly accurate closed-domain transcription tailored for specific customer service environments. With expertise in speech data harvesting and model training, this solution is designed to handle diverse accents, ensuring reliable performance across varied user groups.

NLP-driven knowledge discovery and information understanding

We specialise in research and development of cutting-edge NLP capabilities focused on text-based data. These include tools for information extraction, sentiment and emotion analysis, summarisation, topic modelling and document classification. We apply these techniques to extract actionable insights that support business intelligence, public discourse analysis, and digital transformation.

Large language models and generative AI for local language support

Our research focuses on leveraging large language models (LLMs) and generative AI to accelerate the development and integration of South Africa’s official languages into modern digital ecosystems. We adapt and fine-tune LLMs for low-resource languages using techniques such as in-context learning, prompt engineering, and multilingual training.

Retrieval Augmented Generation and Agentic AI

We build Retrieval-Augmented Generation pipelines and autonomous agentic AI systems that improve efficiencies in tasks such as to content generation, document understanding, expedited information retrieval, closed-domain translation, and customer service.