coventry, uk university of warwick

Josh Beira

Computer Science student and the head of Machine Learning and Software Engineer.

I work on perception and learning systems that have to be right in under 50 milliseconds. Mostly PyTorch, ROS and a lot of coordinate frames. Currently training the models that steer an autonomous race car, and researching text-to-image person re-identification at Warwick.

selected work

04

Text-to-Image Person Re-Identification

Bursary-funded independent research. Built a generative data augmentation pipeline using Stable Diffusion to replace backgrounds and isolate subjects, then fine-tuned CLIP on the noise-reduced set with custom loss functions.

Stable Diffusion CLIP PyTorch
2026

Real-time perception and steering for an autonomous race car

Segmentation CNNs detect track cones; LiDAR clustering lifted accuracy 47%. RL and supervised steering models cut error 34%. Docker-packaged, ≤50 ms from frame to command.

PyTorch OpenCV ROS Docker
2025 —

AI finance assistant — financial literacy and contract analysis

Led a four-person hackathon team. Asynchronous workflows orchestrate parallel Gemini 2.5 Pro agents through the Google API, cutting end-to-end latency. Demoed to non-technical judges.

Flask React Gemini 2.5 Pro SQL
2025

COVID-19 mortality risk model, from scratch

Logistic regression written without high-level ML libraries — NumPy, Pandas and hand-derived gradient descent. Feature-engineered a large clinical dataset to 75.82% accuracy, evaluated on precision and recall.

NumPy Pandas no sklearn
2025

experience

2022 — now
Apr 2026 — Jun 2026

Research Intern, Artificial Intelligence and Machine Learning

Dept. of Computer Science, University of Warwick

Awarded a competitive university bursary for independent research into a more efficient approach to text-to-image person re-identification. Formulated custom loss functions and benchmarked against academic baselines.

Jan 2026 — present

Organiser — Logistics and Outreach

The Warwick Hackathon (WHACK)

Secured £5,000 of sponsorship from leading tech firms, opening the event to 300+ students nationwide. Ran venue, power and logistics with university stakeholders.

Oct 2025 — present

Head of Machine Learning

Warwick Racing

Own the perception and control stack: image-segmentation CNNs with LiDAR clustering, RL and supervised steering models, SQL over large datasets, Docker delivery. Work across the electrical and mechanical teams to keep vision-to-command under 50 ms.

Jun 2022 — Sep 2024

Founder and Lead Developer

Student web development agency

Founded and led a cross-functional team delivering frontend, backend and database work for 10+ local businesses. Owned the full SDLC from client consultation to deployment and maintenance.

Sep 2025 — Jul 2028

BSc (Hons) Computer Science

University of Warwick

Coventry, UK.

Sep 2023 — Jun 2025

A-levels — A*A*A*A*A*

Cadbury Sixth Form, Birmingham

Mathematics, Further Mathematics, Computer Science, Physics, EPQ.

writing

placeholder entries — send real posts and I'll swap them in.

Why LiDAR clustering fixed what a bigger model could not

notes placeholder

Writing logistic regression without sklearn, and what it teaches you

notes placeholder

Fifty milliseconds: budgeting latency in an autonomous stack

notes placeholder

contact

joshbeira@gmail.com

Always interested in opportunities, collaborations, and interesting conversations.