Analytics engineer for semantic layers and metric definitions
A semantic layer is a promise that the same question always returns the same, correct number. I test that promise. My trust harness checks every measure, dimension and filter combination in a star schema against a ground truth computed independently, and it found that a large share of reasonable-looking questions have no single correct answer until someone declares an allocation rule. On the delivery side, through my own consulting practice I was the CEO's data person at Rime, a Saudi edge-AI retail camera startup, where I moved a multi-tenant Snowflake warehouse onto dbt. Earlier, at EY, I built the reporting layer that partners and senior directors used, and at Geidea I built the data models behind settlement and onboarding reporting.
What I've done
- Audited 90 semantic-layer queries; 40 of them (44%) were chasm queries with no single correct answer until an allocation rule is declared.
- Showed one revenue-by-carrier question returning $149M, $136M or $107M depending on how the join was written; only the last matched true revenue.
- Measured fan-out inflating an order-level freight quote by 38.6% once a filter dragged a child table into the join.
- Moved 11 models onto dbt for Rime's multi-tenant Snowflake warehouse, which holds 50+ client schemas; Rime was a client of my consulting practice.
- Built EY Pakistan's in-house reporting layer, used by its partners and senior directors.
- Re-engineered the settlement and onboarding data models at Geidea, which spanned 400K+ POS terminals.
Public work
- A trust harness for a semantic layer: Its write-up also records a mistake in the harness's first version, which tried to compute freight by product category even though freight is quoted once per order.
- Does on-time delivery make customers come back?: A metric definition decided the sign of a causal effect: -0.38 points under one rule, +0.51 under another.
- Marketplace payout reconciliation: A fee computed on tax hit 7,642 orders and passed every reconciliation, because every record shared the same wrong upstream number.
Contact
Email sabasharat.ali@gmail.com, or find me on LinkedIn and GitHub. Full history on the experience page.
Open to roles in New York or remote US.