Daniel Iyalekhue

AI / ML Engineer

Open to full-time roles Birmingham, UK

Hey, I'm Daniel. I build machine learning systems end to end, from the data to the product people actually open. MSc AI at Birmingham City University, with a paper accepted at ICACIN 2026.

What I do

I build machine learning systems end to end. Forecasting, computer vision, tabular models and the explanations behind them, retrieval, and the APIs and dashboards people use them through.

The rest of that path: cleaning and labelling the data, choosing the model that fits rather than the one that impresses, the evaluation that runs before any of it ships, then the interface someone actually opens. I have shipped all of it, including the parts nobody demos.

The habit underneath it is measurement. I don't say a system works until there's a number that says so, and when the number is bad I put that on the page too. Half the value of this portfolio is the results that didn't go my way.

Before AI I spent years in full-stack web work, turning Figma files into things real users hit, so I'm comfortable owning the interface as well as the model. Four co-founded ventures taught me the rest: pricing, pitching, and shipping to a fixed date with nothing to spare.

Flags a permit breach before the reading arrives Forecasting · AquaSense AI Proof ↗ Captions every image, and knows which ones to hand over Computer vision · AccessOps COCO AI Proof ↗ Decisions a regulator can audit, including the unfair ones Explainable ML · Credit-scoring bias audit Proof ↗ A risk manager in front of every order Systems · Multi-venue trading bot Proof ↗ Answers that cite the page they came from Retrieval · DocSage Proof ↗ Picks the model is never allowed to make Recommenders · AURAFIND Proof ↗

Experience

Blockchain Advisors Ltd AI / Software Engineer Intern R&D engineering on a multi-venue crypto trading bot prototype: platform architecture, market data ingestion, the strategy framework and the backtester, delivered in a fixed ten-week window. Jun – Aug 2026

Carril Agency AI / ML Engineer Built an AI reporting pipeline that pulls live Google Ads, Meta Ads and GA4 data through LangChain and the OpenAI API into a FastAPI backend. Also an NLP competitor-analysis tool (BERTopic, spaCy), a RAG content tool grounded in brand guidelines, and client sites including carrilagency.com and brau.ae. Sep 2024 – now

Amdari Data Science Intern Mentored ML projects on churn prediction, HR attrition and industrial predictive maintenance with XGBoost, Random Forest, SHAP and LSTMs, turning model output into business recommendations in Streamlit and Power BI. Aug 2025 – Feb 2026

10Analytics Data Scientist Python ML workflows with Pandas, NumPy and scikit-learn: cleaning, encoding, scaling and exploratory analysis on structured datasets, reported through visualisations and analytical reports. Feb – Aug 2025

Education

MSc Artificial Intelligence Birmingham City University, 2025 – 2026

BSc (Hons) Computer Science, 2:1 Birmingham City University, 2020 – 2023

Certifications IBM SkillsBuild, Google Cloud Skills Boost and Kaggle: AI fundamentals, responsible AI, LLMs, explainability.

Recommendations

Olumide Olaomo Hillsville Farms

"I've had the chance to see Daniel approach challenges with both creativity and consistency, and it's been impressive to watch.

Daniel is a focused, creative, and dependable person to work with. He has a strong technical mindset and is always looking for better ways to solve problems and improve ideas. What stands out most is his ability to combine creativity with practical execution, especially across technology, web development, AI, and digital projects. I'd highly recommend Daniel to anyone looking for someone hardworking, thoughtful, and easy to collaborate with."

Story

Engineer, researcher and co-founder. How I got here, how I work, and what I'm in the middle of.

I started as a software engineer. Full-stack web products for clients: Figma files turned into responsive interfaces, then the APIs, the database work, the performance budgets and the analytics that come with shipping to people who actually use the thing.

Then I moved toward data and AI. Data science at 10Analytics, a mentored programme at Amdari, then applied AI/ML engineering at Carril Agency: LLM tools, RAG systems and NLP pipelines that turn live data into something useful.

The software engineering never went away, it just got more load-bearing. My most recent internship was AI and software engineering on a multi-venue trading bot: platform architecture, market data ingestion, a strategy framework and a backtester, with containers, CI, type checking, metrics and a secret scanner standing in front of every commit.

Now I'm finishing an MSc in Artificial Intelligence at Birmingham City University. With classmates and Beeswift, a Birmingham manufacturer, I co-authored a paper on evaluating grounded enterprise agents that was accepted at ICACIN 2026.

Alongside the engineering I've co-founded four ventures, three of them operating businesses. That's where I learned to price, pitch and ship to a fixed deadline with limited resources.

How I work

Ship the whole thing. Model, evaluation, interface, deployment. The interesting problems live in the seams between those, and you only find them if you own all four. 01

Make it explainable. If I can't say why it works, it isn't done. A credit model that can't defend a rejection is a liability, not a feature. 02

Measure before I claim anything. Evals over vibes. A number I didn't record is a number I'm inventing later. 03

Report what didn't work. A failed experiment is still a result, and the reinforcement-learning run that didn't beat the baseline is on this site for that reason. 04

Build the boring half first. The risk manager before the strategies, the schema before the AI. That order is the only one that survives contact with real money or real users. 05

Ventures

Co-founder, Hillsville Farms, Hillsville Prime and 4Wheels.ng Took each from concept to an operating business, including a dealership model that marketed partner-owned inventory so it never had to hold stock. Three businesses taught me the same lesson three times: the constraint is almost never the technology.

Co-founder and CTO, Nuclii Took the venture through the BCU STEAM Hatchery accelerator, from customer validation to pitching. One of eight picked to pitch.

Relay, through BSEEN A hyperlocal student delivery marketplace. Worked through validation, pricing and trust, and learned that a marketplace with trust unsolved is not a marketplace.

Right now

Shipping LLM reporting pipelines at Carril, over live Google Ads, Meta and GA4 data.

Finishing The MSc, and the camera-ready for the ICACIN paper with Beeswift.

Learning Agent evaluation, and measuring retrieval quality with RAGAS rather than by reading outputs.

Reading Papers on reward hacking, after an RL run on AccessOps quietly optimised the metric instead of the task.

Writing Notes on things that went wrong, because those are the ones worth writing down.

Updated September 2026.

Stack

The tools I work in every day, from the model to the interface.

Languages

Python
TypeScript
JavaScript
SQL
Java
Bash

LLMs and RAG

OpenAI API
LangChain
Ollama
ChromaDB
FAISS
sentence-transformers
BM25
RAGAS
spaCy
BERTopic

ML and data

scikit-learn
XGBoost
LightGBM
PyTorch
TensorFlow / Keras
Pandas
NumPy

Explainable AI

SHAP
LIME
Captum
Integrated Gradients

Computer vision

OpenCV
YOLOv8
MobileNetV2
ResNet
Grad-CAM

Product and infra

Next.js
React
Astro
FastAPI
PostgreSQL
SQLAlchemy
Supabase
Redis

Ops and delivery

Docker
Docker Compose
GitHub Actions
pre-commit
Gitleaks
pytest
Ruff
Black
mypy
Prometheus
Caddy
Vercel
Git

Accessibility