project_
Golan
Golan researches companies, people, and relationships that can affect a credit decision. findings appear connected, and every conclusion links back to the original source so analysts can understand the risk and verify the evidence.
inside the product_
how the system works
from what people experience to what keeps the product running
the platform researches companies, people, and events, then shows the relationships between them. every conclusion includes a traceable link to its source.
an investigation takes multiple steps: deciding what to look for, searching, checking, and changing direction when a lead goes nowhere. I built the coordination behind that work.
Each investigation runs in a separate environment. I contributed to parallel execution and failure handling across investigations.
data access is denied by default and opened only by explicit rules. agents never receive external-service credentials; the platform makes those requests on their behalf.
product in use_
project screens
why it was built this way_
important decisions
the choices that shaped the experience, operations, and product boundaries.
each conclusion needs to show the source
the evidence is part of the experience itself: each finding points to the source document, without depending on a separate report.
many investigations at once
Investigations vary in length and number of steps. They need to run in parallel without a slow job holding up the queue or a failed job stopping the others.
data stays secure even when the code fails
the database denies access by default and releases only the records each person is allowed to view. an application error does not make the data public.
freedom to change the AI model
the product can adopt a better or more economical model without rewriting the entire research flow.
see the other works
