Tan Wee Joe
Penultimate-year Computer Science student at University College London and software engineer at Refractal, an AI security startup. Before that I ran Brigade command systems for the Singapore Army — the same problem I work on now, one layer down: systems that are not allowed to fail.
[Experience]
-
Refractal Software Engineer · London, UK Jul 26 — now 3 mos
- Built full-stack product features as Refractal's early engineering hire, developing Python services and React interfaces for onboarding, configuring, and monitoring AI-agent deployments.
- Developed and tested APIs, SDK integrations, runtime policy controls, and evidence pipelines that evaluate agent actions, persist auditable decisions, and support cloud deployments.
- Python
- React
-
OpenSRE AI Engineer Intern Mar 26 — Jul 26 4 mos
- Built LangGraph-based agent workflows for OpenSRE, turning incident triage, runbooks, and operational tooling into stateful, tool-calling graphs with explicit control flow for production reliability work.
- Connected observability and SRE systems into those pipelines so multi-step remediation stays traceable, repeatable, and easier to extend than ad-hoc scripts during on-call.
- LangGraph
-
PixelPro Studios AI Engineer · Remote Dec 19 — Mar 26 6 yrs 3 mos
- Engineered real-time data ingestion pipelines handling high-throughput event streams from 2000+ concurrent users, with production-grade ETL, feature extraction, and anomaly detection for downstream ML workflows.
- Built and deployed ML-driven automation services that reduced manual overhead by 60% and increased client acquisition by 200%, owning delivery from experimentation to production rollout.
- Designed and built the studio website at pixelprostudios.sg, owning the front end from layout and copy through to deployment.
- Machine Learning
- ETL
-
Singapore Army Command Systems Platoon Commander Jul 23 — May 25 1 yr 10 mos
- Designed PowerShell scripts for real-time status monitoring and automated fault detection, reducing diagnostic time by 75% across critical infrastructure.
- Directed deployment of hyperconverged infrastructure (Windows Server, Nutanix) for Brigade C2 systems with 99.99% uptime and reliability requirements.
- PowerShell
- Windows Server
- Nutanix
-
Government Technology Agency (GovTech) Software Engineer Intern Mar 23 — Jun 23 3 mos
- Built a no-code workflow automation platform using ServiceNow and JavaScript, reducing development cycles and enabling rapid iteration on data-driven processes.
- Translated complex business requirements into scalable architectures using UML, with emphasis on auditability and compliance-aligned documentation.
- ServiceNow
- JavaScript
-
Centre for Strategic Infocomm Technologies (CSIT) Full-Stack Engineer Intern Mar 22 — Aug 22 5 mos
- Built a cloud-native platform using React, TypeScript, Java Spring Boot, MongoDB, and Docker with real-time data pipelines and behavioural analytics, driving a 50% improvement in engagement metrics.
- Designed data models and filtering logic for longitudinal interaction signals, improving data quality for analytics and potential NLP/LLM downstream use cases.
- React
- TypeScript
- Java
- Spring Boot
- MongoDB
- Docker
[Projects]
OpenSRE
Open-source framework for AI SRE agents that automatically investigates production incidents by correlating logs, metrics, and traces, generates root-cause analysis reports, and executes remediations. Integrates with Grafana, Datadog, PagerDuty, and Slack.
Avenir UX
Autonomous web agent framework that evaluates website usability at scale through simulated human interaction and GUI grounding, combining System Usability Scale, Single Ease Questions, and Think Aloud protocols into full UX reports without user studies. Built at UCL Nexus Labs; published at arXiv.
Sentrix
Agentic AI security platform with real-time intent classification, anomaly detection, and automated escalation for safe LLM workflows. UCL AI Engine NVIDIA Hackathon and Stanford ASES Bootcamp winner.
Onflow
AI/ML research and startup: LLM embedding pipelines and synthetic behavioural personas from interaction data, retrieval and semantic similarity for large-scale simulation. Google hackathon winner; white paper; incubated in two London accelerators.
view all projects show fewer
Hackoholics BH25
DSTA BrainHack TIL-AI 2025: multi-domain AI across reinforcement learning, computer vision, ASR, and optical challenges.
treeminspls
Platform for builders to stress-test products using agentic personas and iterate on their sites from simulated user feedback.
anubis
TypeScript project exploring tooling and automation in the development workflow.
Gendash
Dynamically generates interactive dashboards from any API by analysing data and building intelligent visualisations.
[Writing]
[Hackathons]
-
Build: AI Festival 2026 Feb 26
Shipped Sentrix, an agentic security layer for AI swarms, from idea to demo in one day and won the UCL AI Engine track. Read the write-up
-
DSTA BrainHack 2025 Jun 25
Multi-domain AI across reinforcement learning, computer vision, ASR, and optical challenges. View the repo
[Research]
Avenir-UX: Automated UX Evaluation via Simulated Human Web Interaction with GUI GroundingFeb 26
Wee Joe Tan, Zi Rui Lucas Lim, Shashank Durgad, Karim Obegi, Aiden Yiliu Li
An agent system that evaluates website usability through simulated human interaction with GUI grounding, integrating System Usability Scale, Single Ease Questions, and Think Aloud protocols to generate comprehensive UX reports without traditional user studies.
Read the paper