Mojo

This is an AI chat application meant to provision AI services to end users.

Responses are random because the model is under-trained.

Background

I started the Alpha Mind project with a bold goal: to build a foundationally intelligent, agentic application, open to all, but designed with a clear priority to serve as a sovereign technology solution for African nations. The motivation was rooted in a persistent reality: Africa’s severe underrepresentation in the global technology ecosystem.A gap that was true when I began, remains true today, and continues to widen every day.

Before the AI paradigm shift, catching up to legacy technology giants like Google, Apple, or Microsoft felt nearly impossible for emerging markets. But the arrival of modern AI opened a rare, historic window of opportunity. It democratized access to specialized knowledge, accelerated research at an astronomical pace, and gave individual software engineers unprecedented leverage to build. Executed correctly, establishing a world-class technology powerhouse out of Africa no longer seemed out of reach; This is a wave that I intended to catch. I wanted to help lay the foundation for a true, sovereign technological moat for the continent.

The vision was grand. But vision alone doesn't bypass resource constraints.

Mojo became the practical proving ground for that ambition. The mobile client was built natively in Kotlin using Jetpack Compose, communicating with a Python and FastAPI backend backed by PostgreSQL for data persistence.

For the core intelligence layer, I built and trained a custom model from the encoder up on a single NVIDIA RTX 4050 GPU, drawing architectural inspiration from Mamba-3. Due to compute limits, the model is minimally pre-trained, but it stands as a genuine, end-to-end proof of concept. To deliver real-time token streaming to the client, the model is hosted on the backend with inference orchestrated through a combination of Server-Sent Events (SSE) and Dockerized Redis.

While the broader vision evolved amidst hardware realities, the journey was profoundly valuable. It pushed me to bridge cutting-edge research, custom deep learning implementation, and full-stack systems engineering into a functional pipeline. From research to execution, every iteration demanded the complete application of software engineering principles and proved what is possible with persistence and single GPU constraints.

The Web Prototype

This represents the post-phase of the platform's evolution, transitioning the initial mobile app into a full-featured web client designed for corporate and end-user access

Built using Next.js, React, and TypeScript, this web client interfaces directly with the existing backend to showcase cross-platform token streaming and overall system scalability.