Emmanuel Ariyo.
Self-taught Creative Software Developer, AI Systems Engineer, and NLP researcher focused on low-resource African languages, based in Nigeria. I build scalable web applications, production Speech AI and language systems, and automation tools that solve real problems.
Currently a Machine Learning Engineer at Plotweaver, building multilingual Speech AI for Yoruba, Hausa, Igbo, Nigerian Pidgin, and Nigerian English. Founder of Axiveri, my African AI research initiative, where I publish open-source language, speech, and image generation models.
My Story.
I never went to a university for computer science. I never attended a bootcamp. Everything I know about software development, AI systems, and building products, I taught myself.
I started with the basics: HTML, CSS, JavaScript. The free tutorials, the documentation, the trial and error at 2am. No structured curriculum. No professor to ask. Just curiosity and a stubborn refusal to stay confused about anything for long.
What changed everything was shipping my first real project. Not a tutorial clone, not a demo. Something real that other people could use. The feedback loop of building something, watching people use it, and understanding what actually matters became my education.
That same curiosity eventually pulled me into AI and NLP research, specifically the gap in how well language models and speech systems serve African languages. Under Axiveri, the research initiative I founded, I've built and published open-source models addressing that gap directly: Africlaude, an open-source language model series for African contexts; WazobiaVoice, a multilingual text-to-speech and voice-cloning model covering Yoruba, Hausa, Igbo, Nigerian Pidgin, and Nigerian English; NaijaVox, a Nigerian speech recognition model; and AfriVision-Base, an image generation model trained on African imagery.
Today I build production-grade platforms, AI systems, and automation tools used by real users. I published a benchmark on Kaggle that contributes to AI safety research. I built a multi-agent AI platform, a cooperative savings fintech platform, a streaming and academy product, a multilingual Bible companion app, and offline-first PWAs for Nigerian students, all self-taught, all shipped. I'm currently a Machine Learning Engineer at Plotweaver, where I work on production speech AI for African languages day to day.
Being self-taught isn't a gap in my background. It's the most important thing about how I work. I learned to figure things out. I learned to read documentation. I learned that the only thing between where I am and where I want to be is whether I'm willing to sit with the discomfort of not knowing something yet.
What I Do.
Build Websites & Web Apps
Premium websites, SaaS platforms, e-commerce stores, and full-stack applications. From landing pages to complex multi-user systems, built clean, fast, and ready for production.
Engineer AI Systems & Models
AI-powered applications using Groq API, LLaMA, RAG pipelines, and multi-agent architectures. As an NLP researcher, I also build and publish open-source AI models, including Africlaude, WazobiaVoice, and NaijaVox, under Axiveri, my African AI research initiative.
Automate Business Processes
Telegram bots, WhatsApp automation, AI agents, and workflow systems. Lumeo AI handles thousands of Telegram interactions. Microdragon executes complex real-world tasks from natural language.
AI Research.
Beyond building products, I work as an NLP researcher focused on low-resource African languages, publishing open-source AI research under Axiveri, my African AI research initiative.
Africlaude: Language Model Series
Open-source language models built for African contexts. Africlaude-7B is live on HuggingFace. v2 with Yoruba, Igbo, and Hausa support is in development.
View →NaijaVox: Speech Recognition
Nigerian speech recognition (ASR) model series. NaijaVox-V1 supports Yoruba, Hausa, Igbo, Nigerian Pidgin, and Nigerian Accented English.
View →WazobiaVoice: TTS & Voice Cloning
A ~2B-parameter multilingual text-to-speech and zero-shot voice cloning model covering Yoruba, Hausa, Igbo, Nigerian Pidgin, and Nigerian English in one model.
View →AfriVision-Base: Image Generation
A FLUX-based LoRA fine-tune trained on nearly 15,000 captioned images for image generation reflecting African people, places, and contexts.
View →TruthGuard Benchmark
A 120-question confidence calibration benchmark for evaluating the metacognitive abilities of LLMs. Published on Kaggle, open-source under CC BY-SA 4.0.
View →ML Engineering Stack
Jupyter Notebooks, HuggingFace Transformers, Datasets, Gradio, PyTorch, GPU/CUDA training, pipeline filtering, data curation, web scraping for dataset building.
How I Work.
Ship before it's perfect
Real users teach you more in one week than months of planning. I build, ship, and improve from feedback.
Build for real constraints
Most of my work is built for users on slow networks and low-end devices. Those constraints make everything better.
Understand your tools deeply
I don't use what I don't understand. When something breaks at 11pm before a deadline, mystery code is your enemy.
AI as leverage, not replacement
AI eliminates the boring repetitive parts so I can spend energy on architecture, user experience, and the logic that makes a product unique.
Let's build something.
Open to freelance contracts, full-time roles, and interesting collaborations. If you have a project in mind, I'd love to hear about it.