David
Adarkwah

From raw data to autonomous action. I build the infrastructure in between.

AI & Data Engineer · London, UK

About

Raw data is just noise until you engineer it into action.

From my roots in Ghana to building planetary-scale systems in London, my operating philosophy has remained constant: the best architecture brings order to chaos. I don't just train models. I build the infrastructure that allows intelligence to exist reliably in production. Today, I apply this philosophy at Bloomberg to build global climate data utilities.

01Design Principle

Infrastructure must be invisible, fault-tolerant, and infinitely scalable.

02Current Focus

Transitioning from single-prompt LLMs to deterministic multi-agent swarms.

Awards & Recognition

Predictive Modelling09/2021
Zindi Hackathon Winner

Ranked 1st overall. Engineered the best-performing insurance claim prediction model utilizing highly optimized XGBoost ensembles.

Cybersecurity ML05/2023
Zindi Hackathon Runner-up

Ranked 2nd globally. Architected a robust cryptojacking detection classifier achieving top-tier F1 scores on unseen test data.

Knowledge Distribution2021 & 2022
Keynote Tech Speaker

Delivered technical deep-dives on MLOps, scalable data pipelines, and intelligent systems to audiences of 500+ engineers.

Reinforcement Learning11/2023
AWS DeepRacer Finalist

Trained an autonomous racing agent using PPO, placing top 1% in the London regional summit.

Open Source ML02/2024
Outstanding Contributor

Recognized for core contributions to LLM orchestration frameworks, specifically around multi-agent routing.

Trajectory

Started at the raw data layer. Architected real-time credit scoring inference engines that reduced 5-day manual reviews to 2-minute automated decisions. Mastered classical ML, NLP, and statistical rigour.

A great model is useless if it cannot be served reliably. Shifted focus to the infrastructure layer — owning the end-to-end MLOps lifecycle. Built scalable inference APIs on AWS and designed enterprise data pipelines.

Operating at planetary scale. Engineering high-throughput pipelines processing 14M+ global climate data points annually. Architecting autonomous AI agents for RAG-driven data quality investigation.

Fusing deep infrastructure knowledge with frontier AI. Building Residia's AI-driven escrow trust layer for emerging markets, whilst pursuing advanced ML/AI systems research at Georgia Tech.

Projects

Selected Work.

NyamekoBench

A pioneering LLM benchmarking framework evaluating models on African legal, financial, and cultural contexts.

AI Research
2024
NyamekoBench
Overview

Standard LLM benchmarks are dominated by Western-context tasks. Models scoring 90%+ completely fail when asked about Ghanaian property law or West African economic data.

PythonGCPLLM Orchestration

Residia Infrastructure

An AI-driven platform combining secure escrow, verified offerings, and smart matching.

Proptech Escrow
2024
Residia Infrastructure
Overview

Ghana's rental market is a GHS 12B ecosystem plagued by trust issues. Tenants lose millions annually to fake listings, while landlords face extended vacancies.

ReactFastAPIOllamaEscrow

InferenceGateway

High-throughput, unified LLM API gateway engineered in Go with semantic caching.

AI Infrastructure
2023
InferenceGateway
Overview

Production AI applications cannot be tied to a single LLM provider due to rate limits, downtime, and cost variability.

GoRedisAWS ECS

ModelForge

Service-oriented experiment tracking and model registry platform.

MLOps
2023
ModelForge
Overview

Tracking what was tried, what worked, what was deployed, and why becomes unmanageable without infrastructure.

PythonPostgreSQLS3

Let's build something
meaningful.