Open to 2027 roles in SRE, cloud and AI infra
I run cloud platforms in production and benchmark AI infrastructure. DevOps intern at Primustech, Computer Science at NTU.
169 merged PRs on a production IoT platform · 7 case studies with working models · 320 LLM agent runs benchmarked
Now , updated 12 Nov 2026
Final-year Computer Science at NTU, researching what it costs to serve long-running LLM agents, and a DevOps intern at Primustech on its building IoT and AI agent platforms. Graduating June 2027, and open to SRE, DevOps, cloud and AI-infrastructure roles.
Three systems, a chapter each.
Two platforms I run in production at Primustech, and the research behind my final-year project at NTU. Each chapter has what I built, working models of it, and the numbers.
- Primustech
Kubernetes on GKEBuilding IoT platformCollects live data from building systems (sensors, HVAC, meters) and serves the dashboards and alarms facility teams use.169pull requests merged into a production IoT platform6 case studiesOpen the chapter - Primustech
Agents and RAG on Cloud RunAI agent platformTwo AI agents for facility teams: one answers from the IoT platform's live data, the other from company documents.3environments deployed the same way: dev, staging and production1 case studyOpen the chapter - NTU HyScale Lab, final-year project
SGLang on an A100, Google Cloud VMARIES agent benchmarksMaking the lab's open-source agent-serving framework run long agent tasks, then measuring what serving them costs.320long-horizon LLM agent runs benchmarked for my final-year projectresearchOpen the chapter
This site runs like production.
It has an SLO, a deploy pipeline with rollbacks, postmortems and an API of its own. Poke at any of it.
- Live statusUptime against a 99.9% objective, how it ships, and every incident so far.
- What it costsThis month's bill, metered request by request, and what a traffic spike would do to it.
- Go on callTwo simulated incidents to work through, and a postmortem of your own.
- Ask meQuestions answered from this site's own facts, with sources and a trace.
- For agentsThe site is an MCP server: point your own AI agent at it.
- Decisions11 architecture decision records, and what each one costs.
- EvidenceWhere each number on this site comes from, and which you can check.
Latest writing
All posts and postmortems- 11 Nov 2026 · PostRunning an agent benchmark on a supercomputer, without rootARIES started every part of a run with Docker. A supercomputer gives users neither Docker nor root, so I added an Apptainer backend: 6,678 lines, a way to measure containers that have no cgroup, and one hang that looked like the agent's fault.
- 19 Oct 2026 · PostWhat 320 agent runs taught me about serving LLM agentsMCP tool calls were 2.7% of an agent's time, 96% of its input tokens came from the prefix cache, and the GPU sat idle most of the time it was held.
- 19 Oct 2026 · PostFinding a metrics bill in 914 orphaned Kafka topicsManaged Prometheus was the third-largest line item and nobody had measured why. Attribution first, then a reaper, then dropping what nobody read.
Hiring for reliability?
I graduate in June 2027 and I'm looking for SRE, cloud, DevOps and AI infrastructure roles.