Daniel Akhabue

Senior AI Engineer · Senior Data Scientist · Robotics Researcher · Kaggle Expert (#416 Global Rank)

I build intelligent systems that work in production: not just in notebooks. From multi-agent banking AI to computer vision security systems, I bridge cutting-edge AI with real enterprise impact.

I'm a Senior AI Engineer and Data Scientist based in Lagos, Nigeria, with 5+ years of experience building and deploying production-grade AI systems across banking, healthcare, government, EdTech, and Enterprise Systems.

I specialize in the full AI lifecycle: from data engineering and model training to LLM orchestration, multi-agent systems, and cloud-native MLOps on AWS, GCP, and Azure. I don't just build models; I build systems that enterprises trust.

• 🏆 Kaggle Expert: Global Rank #416
• 🥇 4x Hackathon Winner: including 1st Place at Zaka 2023 AI Challenge & Deep Learning IndabaX Nigeria
• 🎙️ International Speaker: PyCon Kenya 2025 · Datafest Africa 2024 · Black in Robotics 2025 (USA)
• 🏫 AI Faculty, Univaciti TeSA Program: training the next generation of production-ready AI engineers
• ☁️ Contributed to Qucoon's AWS AI Competency in Generative AI & Machine Learning Services
• 🌍 Organising Team Member & Faculty, AISOC (AI Summer of Code): the world's largest community for incubating AI talents
• ✍️ I write about AI strategy and engineering at Quantum Reports on Medium
• 🤖 Robotics Researcher: exploring the intersection of AI, perception, and autonomous systems
• ⚡ Fun fact: I'm a certified Data Bender: I make data do things it didn't know it could

My research lies at the intersection of autonomous systems, multi-agent orchestration, and edge intelligence. I focus on the architectural design of communication mesh networks and decentralized coordination protocols for multi-robot swarm systems spanning aerial, terrestrial, and aquatic platforms. A core emphasis is cost-efficiency and fault tolerance through localized machine learning deployed directly on resource-constrained embedded hardware.

I am further interested in federated learning frameworks for collaborative model training across distributed robot nodes, and reinforcement learning approaches for adaptive control and emergent coordination in multi-agent settings. Beyond low-level control and multi-robot path planning, I am interested in distributed Vision-Language Model (VLM) reasoning across robot networks to enable collaborative, real-time spatial awareness and autonomous decision-making in complex, dynamic environments.

Problem: FedAvg is assumed to outperform local training on edge devices, but this assumption is untested in the minimal-client regime (2–3 physical devices) that most real deployments start with
Solution: Ran controlled experiments on real heterogeneous hardware (Raspberry Pi 4 ARM64 + x86_64 gateway) comparing local-only, centralized, and FedAvg on system telemetry anomaly detection — then isolated client count from heterogeneity via a sub-client sweep
Finding: FedAvg fails at 2 clients (0.789 F1 vs. 0.895 centralized), peaks at round 5 then degrades — client count, not data heterogeneity, is the causal variable. Above ~6 clients, FedAvg recovers and wins
Stack: Federated Learning FedAvg Raspberry Pi Edge AI PyTorch Anomaly Detection Python

Problem: Enterprise digital banking overwhelmed by customer enquiries, manual transactions, and fragmented multilingual support
Solution: Pioneered a serverless, multi-agent conversational AI system on AWS Bedrock + Lambda with intelligent routing, stateful context management, and PCI-DSS compliant core banking integration
Impact: Scaled transactional banking by 70% · Cut customer enquiry workloads by 66% · Achieved >99% accuracy with sub-second latency
Stack: AWS Bedrock Lambda RAG Multi-Agent Systems Python

Problem: Traditional surveillance unable to process multi-modal, real-time inputs for comprehensive threat detection
Solution: Architected a CV pipeline fusing drone imagery, satellite feeds, thermal cameras, and ground sensors for 24/7 anomaly and intrusion detection with geospatial analysis
Impact: Enabled border security, infrastructure protection, and remote area surveillance at scale
Stack: PyTorch OpenCV Geospatial AI AWS Deep Learning

Problem: Manual EdTech content development was slow, expensive, and inconsistent across global cohorts
Solution: Built a proprietary AI model for automated course generation using Azure AI, integrated with Agentic student assistants for real-time learning support
Impact: Reduced content development cycles by 95% · Improved query resolution by 60% across 10K+ student interactions
Stack: Azure AI LangChain Agentic AI Python FastAPI

Problem: E-procurement customers faced slow, fragmented data access and poor support experience
Solution: Led design of an Agentic AI support system backed by a 500+ test case evaluation framework covering hallucination detection and tool-use accuracy
Impact: Accelerated data access and purchase information retrieval by 90%
Stack: Python LangChain SQL MLOps Drift Detection

🗂️ See all projects → bit.ly/dan-akhabue

current_focus = {
    "at_work":     ["Multi-Agent AI Systems", "IoT + Computer Vision Fusion", "AWS Bedrock"],
    "research":    [
        "Swarm & Chain Drone Systems: architecture, mesh comms, decentralized coordination",
        "Edge ML on resource-constrained embedded hardware with fault tolerance",
        "Distributed VLM reasoning across swarm networks for real-time spatial awareness",
        "Multi-robot path planning & autonomous decision-making for aerial missions",
        "Deploying intelligent autonomous systems in infrastructure-constrained environments"
    ],
    "teaching":    ["Production AI Engineering", "Cloud-Native MLOps", "Generative AI Systems"],
    "competing":   ["Kaggle", "Zindi Africa"],
    "north_star":  "Is the AI solving the problem it's expected to solve, under acceptable conditions?"
}

• 📄 Does Federated Learning Need a Crowd? Client Count, Not Heterogeneity, Explains FedAvg's Failure on Real Edge Hardware (Best Poster — Deep Learning Indaba 2026, Poster GP-80)
• 📄 My Last Umoja Hack Africa Experience as an Undergrad: The Good, the Bad, and the Ugly
• 📄 Driving Electoral Integrity with Data Visualization: How a Power BI Dashboard Can Revolutionize Your Election Process
• 📄 Is Hacking Together Your First Python Package Difficult?: A guide for data students
• 📄 Becoming a Contributor to the Open-Source Data Community
• 📄 Exploratory Data Analyses: NPFL's Teams and Standings: A novel data mining & EDA project

📚 Read more → Quantum Reports on Medium

Open to consulting engagements, speaking invitations, research collaborations, and conversations about hard AI problems.

📧 akdaniel0009@gmail.com