Entity resolution with Jev cut pipeline costs 99.56% and boosted throughput 7.35x

hrishioa · x · 2026-09-20

Hrishi Olickel (Southbridge.AI) details integrating TypeSafe's system-one model Jev into a high-throughput entity resolution pipeline joining Ohio campaign finance donors and committees. A hank-assisted design cut model costs by 99.56% ($9.83 → $0.3012 on a hard subset, a 32.6× reduction), lifted call-time throughput to 7.35× baseline, and kept quality near-Fable (199/200 families). The architecture batches evidence, routes most batches to cheap models, and escalates only referred hard cases to stronger models with per-call codons. Caveat: wrong signals or prompts cause large mistakes at scale. Prompts and integration principles included.

Original post →

More from coding & agent

coding & agent channel →