Solving African Problems with Deep Tech.

Context-Aware Translation for Low-Resource Languages

The Problem: Large Language Models (LLMs) like GPT-4 often fail to capture the cultural reasoning embedded in African proverbs, resulting in literal but meaningless translations.

Our Solution: We are curating a proprietary dataset of Yoruba and Igbo proverbs annotated with deep cultural context. We fine-tune open-source models (Llama 3, Mistral) using Parameter-Efficient Fine-Tuning (PEFT) to create models that understand meaning, not just syntax.

Current Focus:

• Handling ambiguity in Yoruba tonal marks.
• Benchmarking "Cultural Reasoning" capabilities of standard LLMs.

Offline Plant Disease Detection via Edge Computing

The Problem: Rural farmers in Nigeria lack consistent internet access, making cloud-based AI diagnosis tools useless in the field.

Our Solution: We develop lightweight Convolutional Neural Networks (CNNs) optimized for mobile deployment. By applying model quantization (32-bit to 8-bit), we enable high-accuracy inference directly on low-end Android devices without requiring an internet connection.

DDoS Attack Detection in Multi-UAV Networks

Research Focus: As Unmanned Aerial Vehicles (UAVs) become common in agriculture and surveillance, they become targets for cyberattacks. We are researching machine learning algorithms capable of detecting Distributed Denial of Service (DDoS) attacks in real-time within dynamic, multi-node UAV mesh networks.