I hand repetitive work to reliable systems, giving teams back the hours for work that needs judgement.
{
"name": "Khoa Phung",
"title": "Dr",
"role": "Lecturer in Computer Science, UWE Bristol",
"also": "CTO and co-founder, DKBN World Ltd",
"research": [
"AI automation",
"agentic AI",
"software fault prediction",
"formal methods"
],
"publications": 18,
"citations": 92,
"shipped": 6,
"based_in": "Bristol, UK",
"since": 2016
}I'm a lecturer in Computer Science at the University of the West of England, Bristol, and a Fellow of the Higher Education Academy. I completed my doctorate in 2025, during which I developed error-type metrics: software metrics designed to predict not just whether a program might fail, but what kind of fault it may contain.
My PhD research sat at the intersection of formal methods and machine learning. I used stream X-machines to describe systems as states and labelled transitions, then combined those models with deep learning trained on code to make software fault predictions more specific and actionable. My current research interests include AI agents, multi-agent systems and agentic AI for automation.
I also enjoy turning ideas into working software. Since 2024, I've been CTO and co-founder of DKBN World; before that, I led development on investment and EdTech platforms. I like productive teams, straightforward conversations and sensible engineering. I'm easy-going about most things, although broken builds have been known to test that theory. I moved from Vietnam to Bristol in 2016 and have called it home since.
Numbers, and where they came from.
Production work, led or built end to end. Newest first.
Stedano takes on the weekly admin jobs that small businesses still rebuild by hand every month, so nothing depends on one person remembering where things are. Each use case is built on domain logic from my own practice, and the work leaves behind something the business can actually query later.
a process one person remembers — Stedano → a record the business can query
A SaaS platform for building and selling wedding websites. Template marketplace, live editor with instant preview, RSVP and guest management, AI-assisted content generation, and Stripe checkout, deployed as containers on Google Kubernetes Engine.
Career guidance for students in the Middle East, backed by a scraped and normalised database covering more than 500 regional universities, and an assessment gateway that matches profiles to courses. Led a team of five on a two-week sprint cycle.
A platform connecting businesses in the South West of England with investors, built with partners including Advantage Business Angels, QTEC, Silicon Gorge and Fintech West. Led a team of four and built the authentication and role-based access layer.
Redesigned and untangled MachD's backend, migrating its data layer from Firebase to PostgreSQL to improve performance, simplify operations and align the architecture with the platform's purpose. Integrated the backtesting logic and an AI assistant that reviews results and provides feedback, then worked with the front-end team by defining API endpoints and integration specifications for the UI.
A generator for deliberately messy datasets, used for testing data pipelines and migration tooling. Built independently, IP retained, licensed for commercial distribution.
A tool for building minimal stream X-machine models and generating test cases from them using a state-counting approach, applied to modelling chronic disease progression.
Four strands, current work first. Three of them ask which faults, not how many. The newest asks what a system can safely be left to do on its own.
Which decisions can be handed to agents that plan and coordinate, and what has to hold before a real workload is trusted to them. Applied to automation and to healthcare monitoring.
Applied Sciences 2026 · AI in Healthcare 2026 · IJPEDS 2025
A metric set that classifies faults by type rather than counting them, so a prediction states what is likely to break and not only how much.
IEEE Access 2023 · Software Quality Journal 2025
Deep architectures trained on error-type metrics, including the effect of class imbalance on defect detection.
Concurrency and Computation 2026 · IDC 2024
Stream X-machines describe a system as states and labelled transitions. Used here to generate test cases and to model interacting chronic diseases.
Programming and Computer Software 2021 · CONISOFT 2020, 2021
Undergraduate and master's teaching at UWE and its partner institutions.
Modules I teach include Distributed & Enterprise Software Development, Web Data Applications and Business & Professional Skills.
More than 30 undergraduate and master's dissertation projects supervised since 2020, covering mobile applications, web platforms and applied machine learning.
Link tutor for the BSc Computer Science programme at UWE partner institutions, running pre- and post-moderation and sitting on partnership operations boards.
Fellow of the Higher Education Academy · PGCert in Academic Practice
Open to research collaboration, PhD supervision enquiries and consultancy work on Django and React systems. Reach me through one of the professional profiles below.
