Jason Shen / Trading infrastructure · AI workflows · memory systems
Inspectable systems for trading, AI workflows, and memory under pressure.
I build systems where internal state is not allowed to pretend it is truth.
Trading infrastructure taught me one hard rule: intent, memory, risk, and provider-confirmed truth must stay separate enough to be reviewed. This site is where I make those boundaries visible across trading systems, AI workflows, and memory infrastructure.
Systems
Three systems run on one rule: state must stay reviewable.
Trading Infrastructure
Execution, risk boundaries, reconciliation, monitoring, and provider-confirmed truth.
Status
Trading State Replay is inspectable now.
Open case file →
AgentOS
Durable memory, ephemeral context, human review gates, and filesystem-backed continuity.
Status
Exploration shell; memory model note pending.
Open case file →
AI Workflow Runtime
Intent → plan → tools → verification → artifact handoff.
Status
Case-file shell; workflow trace under construction.
Open case file →
Writing and lab
I write the reasoning down before the playbook exists.
Three surfaces, by how settled the thinking is: Ansatz for field notes, Eigenvalues for essays that have to hold up, Impuls for experiments you can run.
The first essay is live: Maxwell's demon as a sharper metaphor for market efficiency — information cost, and trading edge as state classification.
Collaboration entry
Three kinds of problem where a conversation is worth it.
Best fit:
- trading, execution, or risk systems with unclear state boundaries
- AI workflows that need memory, review, and tool discipline
- technical partner conversations around infra-heavy products
Architecture lens
Every system here runs the same loop.
source → core → validation → feedback