Agentic Research Pipeline · Working Draft
Today the channel picks a topic, then researches it. The goal: research flows into a model continuously, and topics surface when they're ready.
Below: the agents we've designed, the artifacts they've produced, and how sectors move through the pipeline. Every card opens a real document — same files the founder gets to keep.
Took "Pakistan's Untapped Sesame Oil Market" — their first Unlocking Growth episode. Downloaded the audio, transcribed it with Whisper.
Reverse-engineered the script: hook → scale → market → demand → catalyst → value gap → entry → CapEx → OpEx → unit econ → players → personas → CTA. Codified as a JSON schema with explicit data slots.
What kind of data would surface another sesame? Built a trade-data scan filter (PK scale + 5y momentum + raw-vs-processed gap + named entrants). 30+ sectors surveyed; 11 fit the shape.
For each chosen candidate, an 8-section deep-dive fills the same 13 slots as sesame did. Every claim cited. Every unfilled slot becomes an interview question. When all slots fill → episode-ready.
The big shift: instead of "pick a topic, then research it", the system is "research a corpus, watch which topics surface." The sesame template is what defines "surfacing."