Ancestriq
An AI ancestry story generator that turns family records into personalised narratives — where the record layer mattered far more than the model call.
The Archive is a records building: long shelves, indexed drawers, one reading desk. That is the honest shape of this project.
Ancestriq turns family records into personalised ancestry narratives — you give it what is known about a lineage, and it produces a readable story rather than a table. The interesting part was never the generation. It was the records.
The actual problem
A generative model handed loose family records will confidently produce a plausible narrative that is not in the records. For anything genealogical, a plausible answer is worse than no answer, because an invented ancestor reads exactly like a real one — and this is someone’s family.
So the work went into the layer underneath: getting records into a structure where every claim in the output traces back to a specific input, and where the model’s job is narrowed to phrasing what it was given rather than deciding what is true. If a detail is not in the records, the story does not get to have it.
What I learned that carried forward
Two things, and both of them are visible elsewhere in this city.
The first is that the record layer is the product. Swapping the model is an afternoon. Getting the data into a shape where the output is constrained by it is the project. I have not seen a GenAI application where this was not true.
The second is knowing when not to reach for a model. That lesson is the direct reason this portfolio contains no AI at all — no chatbot, no assistant, no inference at runtime. Every question a recruiter might ask is answered by authored text in the Dossier, and every behaviour in the game is deterministic logic. Having built with these models, I have a clear view of where they earn their latency and their bill. A portfolio is not one of those places.