about
Human-aligned, Machine-verified
logitly builds agentic AI systems that write data humans can trust: agent traces, tool-use dialogs, and evaluation sets.
01 / why we started
access
The frontier is fenced. In much of the world, people cannot use AI on equal terms.
recipe
Open weights, closed data. The model is shared, the ability to improve it is not.
02 / our approach
generate
No annotators. Written by agents: every trace, dialog, and evaluation.
verify
No reviewers. Machine verified as a human would expect.
grow
No experts. Built to scale, hungry for GPUs.
03 / what we offer
system-one-raw
Is an answer supported by its state?
flash-coding-sft
Coding agents that reason and use tools.
flash-coding-dpo
Coding agents that complete or fail.
04 / founder
founder & CTO
PhD in CSE, CUHK. Works on GPU efficiency, LLM training pipelines, agent loops, tool use, and post-training for search and coding agents.
now
Scaling to larger tasks and multiple modalities: longer agent traces, richer tool environments, and domains beyond coding. The verification bar stays the same.