Agent tooling
LoopPrism
A local-first inspector for coding agent sessions. Explore execution, context, timing, and estimated API costs across tools such as Claude Code and Codex.
Thoughtful tools. Useful AI.
Building AI systems
that act reliably.
I build tools for agent execution, evaluation, and failure recovery. Clear boundaries, visible actions, and results people can check.
Previously at Kuaishou and FanRuan, working on agent platforms, RAG, and backend systems.
Selected work
01 / BuildModel engineering
nano-posttrain
A compact PyTorch implementation of SFT, DPO, and LoRA. A learning reference for exploring post-training.
Product exploration
OfferPilot
A project exploring tools to help graduates navigate their job search.
AI systems · Course project
TACC AgentForge
An in-progress course project exploring model serving and a tool-using shopping advisor. Initial model deployment and smoke tests are complete.
Let's build something useful.
02 / ConnectHave an agent workflow to improve, an open-source idea, or a question? I'd be glad to hear from you.
markfuture.ai@gmail.com