Branson Tay

Full-stack engineer · forward-deployed

I build applied-AI
tools that ship.

A full-stack engineer who ships production code end-to-end, with tools living on the internet and at the race track with real users. I drop into unfamiliar systems and love using entrepreneurial instincts to own problems from scoping to deployment.

OPEN TO WORK LONDON-BASED US-AUTHORISED (H1B1) FROM SEP 2026

PRJ/001 FEATURED

Helm

Automatic CV generator app, built with an agentic harness on a self-improving loop.

  • Live consumer product 10 users 20 CVs generated
  • Generates a tailored CV from one job description
  • Uses end-to-end evals and agentic harness to determine the quality of the CV output

WHY Teach myself system design, and self-improving harnesses in production environments

Source is private as Helm is a live product. The diagram shows the architecture instead.

  • typescript
  • fastapi
  • k8s
  • sqs
  • cloudflare
HELM · SYSTEM TOPOLOGY
k3s · EC2 (t3.small)AWSEXTERNAL SaaSHTTPSTraefik ingresssession APIPOST /pipeline/runOAuth 2.0checkout · webhookspublish jobusers · sessionsbullet cacheconsumeSonnet · HaikuPUT cv.docxBrowserclientCloudflareDNS · CDN · WAFweb:3000Next.js · trust boundaryapi:8000FastAPIworker:9000Python · pipelinePostgresshared stateSQSasync jobsS3artifactsGoogleOAuth 2.0Claude APISonnet · HaikuLemon Squeezybilling
k3s cluster · EC2AWS managedexternal SaaS

Helm exposes one public surface (web) and keeps every other service private behind a shared key. Select any component to see what it does, and why it sits where it does.

PRJ/002 RESEARCH

MARL Racing Sim

When teamwork helps and when it hurts.

  • Multi-Agent Reinforcement Learning in a custom-built F1-inspired action space
  • Find learning strategies for teammates in a dynamic competitive environment
  • Maps when team play speeds learning, and when it stalls it

WHY Investigate when cooperation and competition hurt AI agents working together.

  • python
  • pytorch
  • dqn family

PHASE 01 · DOES IT LEARN?

Purpose: Let's verify if the simulator works as an experimentation platform, and DQN (Deep-Q-Network) learns. One vanilla DQN against a base agent with simple heuristics.
Conclusion: All three seeds for our DQN agents beat the base agent more than 50% of the time. We got DQN to work!

Agent 1Base agent
PRJ/003 SHIPPING

Quasi-Static Lap Time Simulator

Lap-time solver built from first principles, with a fitted tyre model

  • Actively used by Leeds Gryphon Racing to determine design trade-offs
  • Predicts pace per track from mass, aero, tyres, and gearing
  • Python FastAPI web app for engineers to easily use with RAG AI chatbot

WHY Let engineers reason about design choices in minutes, instead of hours of expensive track time

  • python
  • fastapi
  • api design
  • linear programming
  • web development
  • data modelling
LAP TIME →
MASS →
LAP TIMEs67.063baseline
TOP SPEEDkm/h113.9
MAX LONGITUDINAL Gg1.531
MAX LATERAL Gg1.3125
MORE WORK

Other projects

I like transforming new technologies into products that translate into a breadth of real-world applications

  • Aerodynamic Concept RAG Analyser A RAG pipeline that ingests arXiv aerodynamics papers into a vector database for LLM evaluation. pythonragvector dbllmmcpfastapi
    • Assembled a RAG pipeline ingesting arXiv aerodynamics papers into a vector database for LLM evaluation
    • Integrated MCP tooling to extend agent retrieval, enabling dynamic paper ingestion and concept refinement
  • Travelindr Co-founder and CEO of a group travel-itinerary generator, taken from idea to an incubated MVP. agilea/b testingproduct strategyfundraising
    • Founded Travelindr, a group travel-itinerary generator, in a small team implementing Agile practices
    • Communicated core company strategy to the team, backed by qualitative focus group data using A/B testing
    • Pitched to investors, reached Minimum Viable Product (MVP) status, and was incubated at NEC
  • Azure Serverless Distributed Pipeline An event-driven distributed pipeline on Azure for serverless analysis at scale. pythonazure functionsblobqueueserverless
    • Designed a distributed Azure pipeline using Functions, Blob and Queue Storage for event-driven analysis
    • Implemented serverless functions and load-tested HTTP triggers to validate throughput and latency at scale
  • E-commerce Shopify Store Built and marketed e-commerce brands, optimising conversion from data and heatmaps. shopifyfacebook adsphotoshoppremiere pro
    • Increased conversions by 10 percent by adapting the add-to-cart button position from cursor heatmaps
    • Built an efficient sales funnel from data insights and trend targeting
    • Developed and marketed multiple store brands with attention-grabbing creatives
  • Formula Student ICE Data Acquisition System A real-time GPS telemetry pipeline from Arduino Teensy to live engineering dashboards. c++arduinogit
    • Built a real-time GPS telemetry pipeline from Arduino Teensy to dashboard and wireless engineering displays
    • Honed collaborative software practice with Git and code reviews
  • 2D CFD Radiator Optimisation Simulator A 2D incompressible fluid solver in Rust that optimises radiator angle for cooling and drag. rustnumerical methodsparallelism
    • Developed a 2D incompressible fluid solver analysing radiator angle effects on cooling, pressure drop and drag
    • Automated batch simulations and visualisations to optimise radiator design with parallel processing
PROFILEWHO I AM

Background

I sit at the intersection of mechanical engineering and computer science, and I build in that overlap. I'm drawn to where AI meets the physical world, where learning systems meet real hardware and dynamics.

Experience
  1. 2024 – 25
    Jaguar TCS Racing Junior Strategy & Software Engineer Kidlington, UK
  2. 2020 – 22
    Republic of Singapore Navy Marine Systems Specialist Technician Singapore
Education
University of Leeds2022 – 26
BSc Computer Science, Year in Industry · Leeds, UK
  • Predicted first-class
  • Formula Student Performance & Simulation sub-team lead
Ngee Ann Polytechnic2017 – 20
Diploma, Mechanical Engineering · Singapore
  • Designed gearbox to load, speed and dimensional spec in CAD
  • Graduated with Good Progress and School of Engineering Merit awards
Stack
LanguagesPython, TypeScript, C++, Rust, MATLAB, Java
AI / MLPyTorch, RAG, Claude Code, Codex, agentic harnesses, NumPy / SciPy
Web / InfraReact, Node, FastAPI / Flask, MongoDB, AWS, Azure, Kubernetes, CI/CD
ToolsGit, Docker, Azure DevOps, Plotly

Let's build something
that ships.

Download CV
Looking for
Roles
Forward-Deployed EngineerApplied AI EngineerSolutions ArchitectFull-stack Engineer
Locations
London51.5°N · 0.1°W
New York40.7°N · 74.0°W

London-based UK Graduate visa US-authorised via H1B1 treaty visa

Available
SEP 2026
GitHub LinkedIn © 2026 Branson Tay