Building production AI systems for public health, NGO/UN, and EdTech — where reliability and privacy matter as much as intelligence.
I design systems around three constraints: minimal dependencies, privacy by default, and first-principles thinking. In practice, that means offline-capable tools where connectivity can't be guaranteed, deterministic fallbacks where AI shouldn't be the only safety net, and architecture decisions I can defend from first principles — not because a framework made the choice for me.
Co-founder equity in Algorythyme and Pedagic Distribution Nigeria Ltd. ~1.5 years shipping production AI across Algorythyme, AI4AI, Mensah Philosophic Circle, and Zummit Africa.
|
🏥 Offline CHW Clinical Decision Support Tool Gemma 3 1B (Q4_K_M, via llama.cpp) driving a deterministic clinical rule engine for community health workers with no reliable connectivity — with a hard emergency-bypass path so the AI is never the only thing standing between a patient and urgent care.
|
📊 DHIS2 Public Health Intelligence Assistant FastAPI + LangGraph + pgvector system that turns raw DHIS2 public health data into queryable, multi-modal insight, with dual-mode authentication for institutional deployment.
|
|
🛡️ Gemma Sentinel Fine-tuned Gemma deployment for monitoring and inference — small model, tightly scoped, production-hardened.
|
🔌 IntelliPlug LLM-driven logic layered onto hardware control — bringing agentic decision-making to a physical automation project.
|
|
📝 Sovereign Minute Maker Lightweight, privacy-first meeting-minutes generator — no data leaves the machine that doesn't need to.
|
🎯 Job Market Intelligence Agent Agent-based discovery system that surfaces and extracts signal from job market data at scale.
|
Focus areas: Retrieval-augmented generation · Multi-agent orchestration · Model fine-tuning (Gemma, Unsloth) · Offline-first AI deployment · Privacy-preserving system design