Welcome
Welcome
Welcome
Welcome
AI engineer.
Deep in the stack, close to the customer.
Scroll ↓
Basically, I do
AI Product Engineering
AI Inference Engineering
Backend Engineering
Solution Architecture
Nice to meet you — let’s chat01 — Work
Internships, research and open source — scroll sideways.
Solution Architect Intern
A serverless GenAI workflow that auto-generates personalized marketing sites for exec-level clients — now used across the London AWS account-manager team.
2-week job → 5 min
Backend Engineer Intern
A full-stack cost tool, built from scratch, that finds idle and underused resources across the company's GCP and AWS.
flagged $18k/mo in waste
Edinburgh BSc Dissertation · Distinction
Extended the ServerlessLLM (OSDI ’24) framework: prefetch weights over GPU DMA and model switching as a two-machine flow-shop — higher GPU utilization, lower switch overhead.
Makespan 66× faster
Personal · live ↗
A site for finding premium London Build-to-Rent — 300+ listings with commute routes, nearby stations and building amenities (gym / pool), plus an embedded AI property agent (tool calling + ReAct).
Real users · completion 30% → 95%
Open-source · via MoE-CAP
Per-forward-pass expert-activation profiling for Qwen3-MoE — which experts fire, their counts and ratios — for easy MoE benchmarking.
Qwen3-MoE merged
03 — Skills
02 — About
jingsheng.c04@gmail.com
You can call me Jingsheng, or just Jason. I did my Bachelor's degree in Computer Science at Edinburgh, graduated with first-class honours, and I’m now doing an MSc in AI at Imperial. I’ve been a traditional engineer at Thought Machine (a unicorn fintech company), shipping production full-stack systems; did my dissertation on large language model inference under Prof. Luo Mai, graded Distinction; and built reliable LLM agents end-to-end on my own projects. I've also been a solution architect at AWS, I built an AI workflow — proven in real use to cut a two-week job to minutes — and worked with clients directly across discovery calls and workshops, mapping their pain points to solutions.
I’d rather not be confined to a single layer. The strongest AI work comes from people who understand the metal, the model, and the customer in front of it. Deeply technical, and close to who it’s for. That’s what I’m building toward.