I am a Product engineer. For six-plus years I've built B2B SaaS and data-heavy products with remote teams, and lately I've found the work I enjoy most on the AI side: computer-vision pipelines that feed real dashboards, tool-calling and MCP servers, and agents that earn a person's trust one decision at a time. I like building things people lean on every day, and I like getting a little better at this every week.
I'm comfortable in Python (FastAPI), Node (NestJS, Express), Java (Spring Boot) and .NET on the back end, React/Next.js and Angular on the frontend, Postgres and event-driven plumbing in between, deployed on AWS, GCP or Azure with GitHub Actions. I care less about which stack than about whether the thing holds up when a real person leans on it, and I try to write READMEs that say what is not finished as clearly as what is.
I'm still relatively new to building with language models, and that is the part I most want to get better at. The projects below are how I'm doing that in public.
ShelfSense — retail and cold-chain operations for small distributors in Lagos. A field agent photographs a shelf, the system reads it, drafts the reorder, and holds the bigger calls for a manager to approve. Fridges stream temperatures; a sustained fault becomes a technician proposal, also held for approval. Multi-tenant Postgres with row-level security, a 100-case eval set replayed in CI with a regression gate, every model call traced and priced. Python, FastAPI, Next.js, Claude, MCP, Langfuse
SafeGate — a content-safety gateway that sits in front of an LLM. Each prompt goes through classification, policy and response stages, then is blocked, rewritten or allowed, and every decision lands in an audit trail. Live · FastAPI, Angular, Claude
TaskFlow — a Kanban task manager with drag-and-drop. Small, finished and deployed; I built it to get properly comfortable with Angular signals. Live · Angular, TypeScript
- curate — ranks the labels in a detection dataset most worth a human's attention, then measures whether fixing them moved the model.
- ragbench — a harness for measuring how RAG pipeline choices change answer quality, latency and token usage, with confidence intervals.
- expense-tracker-api — a small, complete REST API in ASP.NET Core with EF Core, Postgres and JWT auth.
- Medium
- Comparative analysis of AI-based search algorithms in solving 8 puzzle problems — Bulletin of the National Research Centre (SpringerOpen), 2024. Co-authored during my computer science degree. I'm now pursuing a master's in computer science with a research focus on artificial intelligence.
Open to senior full-stack and AI engineering roles with a team building something hard, remote or with relocation.