Engineering practice · est. 2017

We join platforms during the months the graphs stop making sense.

Goscale Team is a small group of staff-level engineers who embed with product teams when latency tails fan out, write-amplification starts to bite, and the on-call rotation stops sleeping. We work on capacity, reliability, and architectural decompression — the unglamorous middle phase between product-market fit and real scale.

42engagements completed

9industries served

6principals, no juniors

Selected engagements

Names and identifying details are omitted; case summaries are published with client consent after a 12-month cool-down.

Ledger infra · fintech

Double-entry ledger crossed 10⁹ rows without downtime

Replaced a single-writer Postgres ledger with a sharded, idempotent append log; reduced p99 posting latency from 2.8 s to 140 ms.

Search · marketplace

Relevance rewrite without the 9-month relevance rewrite

Incremental move from a single-node inverted index to a tiered retrieval pipeline. Recall@20 held, infra cost dropped 61%.

Queueing · logistics SaaS

Work-stealing scheduler replaced a 40k-line cron monolith

Designed a small queueing substrate on top of their existing relational store. Removed three classes of duplicate-delivery bugs permanently.

Multi-tenant isolation · B2B SaaS

A noisy-neighbor outage at 2 AM became a capacity dashboard by quarter-end

Introduced per-tenant quotas at the query layer and a shadow-traffic harness. Large-customer incidents dropped from weekly to none observed.

Data migration · healthtech

Zero-downtime schema migration across 14 months of dual-writes

Designed the read path, the write path, and the rollback path — then wrote the tooling the client used to run the cutover themselves.

Observability · developer tools

From 400 GB/day of log spam to a three-metric SLO dashboard

Reworked the telemetry pipeline around cardinality budgets. On-call paging volume fell to a quarter of its previous rate within a month.

Additional engagements — in advertising attribution, scientific computing, and government data pipelines — are available on request through an existing relationship.

“The graphs stopped lying about six weeks in. They didn't just fix the hot path; they taught our senior engineers how to see the system the way they saw it.”
— VP Engineering, logistics SaaS engagement, 2023