I am interested in systems that turn confusion into a decision.
That is the same job whether the input is a messy operation or a birth chart. Gather what matters, show the reasoning, and leave the choice with the person who has to live with it.
replace before launch
Three layers, and none of them work alone
People usually arrive for one of these and stay because of the other two. It is the combination that is unusual, not any single part of it.
(01)
Product manager
Years of shipping operational systems that real teams had to use on a Monday morning. This is where knowing what to build comes from.
(02)
AI engineer
Agent workflows, evaluations, and the unglamorous work of making a system behave the same way twice. Currently at an asset manager.
(03)
紫微斗数 practitioner
A year of study and real consultations, checked against colleagues. The domain that forced me to care about traceable reasoning.
Not a straight line, and that is the useful part
Until 2025
Product management
Shipping internal systems. Learning that the hard part is rarely the software, it is deciding what the software should be certain about.
2025
紫微斗数, seriously
Started studying properly, took real consultations, and hit the wall that started everything else: I could not check my own reasoning.
2026
Building the engine
Turned the practice into a system that traces claims. Six months of it, in public, with the failures written down.
Now
AI engineer, and still building
Agent workflows and evaluation by day. The engine and the writing continue, and each one keeps teaching the other.
2027
Advising firms on the same problemplanned
Helping teams put AI where it removes repetitive work without giving up control. The same question as the engine, with a budget attached.
2028
An app anyone can read their chart inplanned
One birth time, several Chinese systems, one coherent reading that shows where each claim came from. The reason the engine exists at all.
2029
八字 to the same standardplanned
Bazi properly rather than casually, so a reading can draw on two systems that actually agree with each other.
What I am good for
- Agent systems where the reasoning has to be checkable
- Evaluations, and deciding what "good" even means
- Sitting with a team and building the thing rather than advising on it
- Turning a messy process into something with a defined output
What I am not
- A prompt consultant. Prompts are not the hard part
- Someone who will promise a model is reliable without evidence
- An agency. It is me, which is the point and also the limit
- Available for work that removes the human from a consequential decision
I do not predict deaths, diagnose illness, or tell anyone whether to leave a marriage. A reading is for reflection. It is not medical, legal or financial advice, and it never replaces a professional who is qualified to give it.
Tell me what you are working on
I read everything that arrives, and I answer the ones I have something useful to say about.