MY WORK
The engineering behind the philosophy.
Everything I teach about productivity and AI is grounded in systems I actually build and ship — multi-agent architectures, retrieval pipelines, and assistants that are reliable in production, not just impressive in demos.
EXPERIENCE
Where the work happens.
AI Engineer · AwaDoc
Apr 2025 – Present · Remote · Full-timeArchitecting and scaling a conversational health AI platform serving 100,000+ users across 20 African countries.
- ✦Architected a LangGraph supervisor system routing across 8+ specialized sub-agents with async store and checkpointing for resilient orchestration.
- ✦Built a production RAG pipeline with Qdrant plus a Firebase pointer-hydration layer to avoid storing sensitive document payloads in the vector index.
- ✦Implemented dual-layer observability with Langfuse traces and Prometheus metrics (invocations, latency, tokens, active runs, errors).
- ✦Delivered multimodal pipelines for voice (Whisper STT), TTS, image analysis (GPT-4o Vision), and PDF/DOCX ingestion with OCR-aware routing.
Engineering Lead · KikeAI
2025 – 2026 · RemoteLed a cross-functional AI, backend, and IoT team building a WhatsApp-native assistant for cooking gas management.
- ✦Designed and deployed a supervisor-worker multi-agent architecture handling tracking, forecasting, ordering, and planning workflows.
- ✦Engineered WhatsApp Business API integration with a FastAPI backend and production VPS deployment using Docker.
- ✦Owned end-to-end reliability and technical delivery across agent workflows, IoT signals, and backend APIs.
AI Engineer (Contract) · ARC Solution
Apr 2025 – Present · Remote · ContractBuilt autonomous content and client-engagement systems with human-in-the-loop control and resilient workflow execution.
- ✦Built a LangGraph StateGraph autonomous SEO pipeline with typed nodes, conditional routing, and PostgresSaver checkpointing.
- ✦Implemented human-in-the-loop approval via LangGraph interrupts and Telegram Bot commands to pause/resume runs asynchronously.
- ✦Shipped a Dialogflow-powered Telegram chatbot for client enquiries and quotes, reducing manual sales-response overhead.
Founder & Productivity Coach · Kairos
5 years · OngoingMy community of productivity-minded people — where the coaching, workshops, and products on this site live.
- ✦Coached career professionals and students 1:1 on mindset, decision-making, and productivity systems for 5 years.
- ✦Built and run a community of productivity-minded people learning to work intentionally — with and without AI.
- ✦Host Top 10% Protocol, a monthly virtual peer meeting for people across fields, and built digital products: the Kairos Journal and Planner.
PROJECTS
Built end to end — architecture to deployment.
Skedi
Client workMulti-agent academic assistant for university students
A conversational Telegram assistant for Covenant University students that handles timetable navigation, handbook Q&A through RAG, and reminder management. Built with a LangGraph supervisor and five sub-agents, GPT-4o structured outputs with vision fallback for course parsing, and async FastAPI webhook processing with PostgreSQL checkpointing for persistent user context.
WAYA
PersonalMulti-agent Telegram academic assistant
A multi-agent Telegram bot that acts as a full academic assistant. Handles timetable management, answers handbook questions via RAG, tracks tasks, and sends broadcast messages to student groups. Built end to end from architecture to deployment.
Trailr
Client workAI career path recommendation system
Hybrid recommendation engine that maps users to career paths using collaborative filtering and content-based filtering. TF-IDF and SVD power the matching logic. Fully deployed on Railway with a FastAPI backend and PostgreSQL database.
Scholr
Client workCampus peer tutoring platform
Platform that connects students with peer tutors using TF-IDF cosine similarity matching. Students post subjects they need help with, the system surfaces the best-fit tutors. Next.js frontend, FastAPI backend, deployed on AWS EC2.
NutriMatch
Client workHybrid nutrition and lifestyle recommendation system
Personalized nutrition engine built on a hybrid collaborative and content-based filtering approach. Trained on USDA and Food.com datasets with Nigerian foods included. scikit-learn TruncatedSVD powers the collaborative filtering layer.
MealBot
Client workLangGraph Telegram meal planning bot
Telegram bot that generates personalized meal plans for users based on cafeteria availability. Uses TOPSIS algorithm for meal scoring and ranking, GPT-4o-mini for conversational interaction, and LangGraph for agent orchestration.
THE TOOLBOX
Skills that ship.
AI & LLM Systems
Data & Backend
Frontend & UI
Infrastructure & Observability
TEAMS & ORGANISATIONS I'VE WORKED WITH
Want your team to work
more intentionally?
Team training, speaking, consultation, or personal 1:1 productivity — it starts with a conversation.
Book a Call