rahim@stdin ~ %
ls projects/
14 entries, 3 to start with
start here
| name | result | kind |
|---|---|---|
| retail-pos | 29 processors across 14 retail partner feeds | work |
| sap | 90 SAP objects, 400+ flows, daily to hourly | work |
| events | first endpoint to queryable table in under four months | work |
production · one AWS platform at Infocepts
| name | result |
|---|---|
| ingest | 50% lower end-to-end processing time |
| warehouse | 35% faster queries behind one serving layer |
| platform | 40% less manual effort per deployment |
| catalog | 154 tables and 5,000+ columns declared in Terraform |
| planning | 10 Glue jobs; a bad file fails the run and names itself |
| flowctl | 400+ flows suspended or rescheduled in one invocation |
| snapshots | 9 change-record tables to one current row per key, daily |
independent · source linked
| name | result |
|---|---|
| warehouse-gpt | context took the smallest model from 63.5% to 98.0% |
| leakproof | ROC-AUC 0.999 offline, 0.901 live; honest held 0.947 |
| lossproof | 69% of the careless lake wrong; the correct one matched |
| nightshift | a 101-line runbook beat the on-call agent, 15 to 11 |
cat one for the write-up: the problem, the architecture, what broke, and what changed.
grep <term> searches every one of them.