About Brilla AI

The Vision

The National Science and Maths Quiz (NSMQ) is a yearly competition that captivates the nation — a contest of intellect, recall, and reasoning among Ghana’s brightest students. It represents not only a measure of academic strength but also a celebration of Ghanaian curiosity and scientific literacy.

Brilla AI seeks to develop an autonomous AI contestant that can hear, understand, and answer NSMQ questions live on stage, under the same time constraints as human competitors. The project’s long-term goal is to demonstrate the possibility of building world-class, contextually grounded AI systems in Africa.

The Challenge

Competing in the NSMQ requires multimodal intelligence. Contestants must process spoken questions in real time, comprehend specialized scientific and mathematical language, and respond with accurate, structured answers — often within seconds. The challenge for AI is to emulate these human cognitive processes through machine learning pipelines.

Brilla AI focuses on the five NSMQ rounds: Fundamentals, Speed Race, Problem of the Day, True or False,and Riddles. Each round tests different forms of intelligence — from factual recall to abstract reasoning and verbal agility. The “Riddles” round, which demands sequential clue integration and hypothesis generation, has become the project’s first live testbed.

The Technology

Brilla AI combines speech, language, and reasoning systems into a unified AI pipeline:

• Speech-to-Text (STT): Real-time transcription of quiz audio using fine-tuned Whisper and Wav2Vec2 models trained on Ghanaian-accented English.
• Question Extraction (QE): Detection of question boundaries, preambles, and riddle clues through semantic segmentation and transformer-based classifiers.
• Question Answering (QA): Context-driven reasoning with models for science and mathematics problems.
• Text-to-Speech (TTS): Generation of expressive, natural Ghanaian-accented speech for AI responses.
• Control Orchestration: Coordinates the system, handling speech segmentation, pause detection, and streaming responses to a web interface.

This modular design mirrors human cognitive function — listening, understanding, reasoning, and responding — but at machine speed.

Building for African Contexts

A key insight from the Brilla AI research is that AI models trained solely on global datasets struggle with culturally grounded, domain-specific contexts. Scientific questions in the NSMQ often include local accents, or analogies unfamiliar to large pre-trained models.

Brilla AI integrates local data collection, contextual annotation, and Ghanaian linguistic variation into its training pipeline. The aim is not only to win a quiz, but to reshape how AI understands and serves low-resource communities.

Research and Open Collaboration

Brilla AI is a multidisciplinary effort — combining computer science, education, linguistics, and social innovation. It encourages contributions from students, educators, and developers who want to help design AI that reflects local realities.

The project’s architecture and data interfaces are open for collaboration with ongoing work on conversational timing, confidence calibration, and explainable reasoning.

Transforming Education

Beyond competition, Brilla AI represents a vision for personalized, adaptive education in Africa. An AI that can think and reason like a top student could eventually become an intelligent tutor — guiding learners through STEM concepts with empathy, speed, and precision.

By bridging local knowledge with frontier AI research, Brilla AI demonstrates that innovation can be homegrown, culturally grounded, and globally competitive.