Short posts on the work behind the mentorship — the plan generator, the benchmark grounding, the process we hold ourselves to.
An architecture write-up of the mentee intake pipeline — hybrid retrieval over an O*NET-style benchmark catalogue, versioned prompts, structured LLM analysis, and a mandatory human checkpoint before any plan reaches a mentee.
How we stopped guessing at career plans — grounding them in an O*NET-style benchmark catalogue with a semantic fallback, wiring CV and LinkedIn intake into the same path, and adding a one-time 24-hour reminder for signups who never finished the workflow.