Notes from the Saytu team

What we learn building a Live2D character that answers preorder, sizing and authenticity questions on anime storefronts.

Five near-matches kept the exact product away from the model

The exact product was already in Shopify, but five broad search results kept it out of the model's context. We moved the hard limit after ranking instead of before it.

Build logSeptember 23, 20266 min read

The AI started talking again. The human request stayed open.

A handoff nobody picked up used to silence the assistant forever. We now let the AI resume after staff silence without clearing the human request, because helping again is not the same as resolving it.

Build logSeptember 23, 20265 min read

Twenty chunks did not mean we had the whole document

A source with twenty stored chunks could mean either complete or truncated. We started recording the pre-cap count so cut, complete and unknown became three measurable states instead of one guess.

Build logSeptember 23, 20266 min read

Semantic retrieval was off in production, and the fallback hid it

Semantic retrieval had been failing with 401s while keyword fallback kept the assistant looking healthy. The first embeddings backfill exposed that production was authenticating with a literal environment-variable name.

Build logSeptember 23, 20265 min read

Right Match Wrong Image

The product match was right; the card still made it look wrong

A shopper asked for white, and we found a product that really had a white variant. But its main card image was black. We changed the rule so a variant-only match must be stated in the reply or the card stays off the rail.

Build logSeptember 22, 20264 min read

Nine tied rows were enough to make the same question retrieve different sources

Nine tied policy rows produced two different top-threes on consecutive calls. The fix was not a better ranking — it was making the ranking we already had give the same answer twice.

Build logSeptember 21, 20264 min read