M. Halberstam
Distributed data · founding principal
Seventeen years on transaction processing, replication, and the parts of the CAP theorem that nobody enjoys discussing at 4 AM. Leads most of our ledger and migration engagements.
Practice
We operate like a surgical team, not a staffing agency.
How we work
A typical engagement runs twelve to twenty weeks. Two principals embed part-time with your senior engineers; a third principal reviews architectural decisions asynchronously. We do not offer a bench, we do not grow headcount, and we do not subcontract. The maximum number of concurrent engagements the practice has ever run is three.
We write code. We review code. We also write memos — short, specific, dated documents that capture tradeoffs at the moment they are made, because in our experience the most expensive bugs in a scaling system are architectural decisions nobody remembers making.
The shape of a scaling event
Most clients call when they have just crossed, or are about to cross, one of three thresholds: the first time a single database stops fitting on a single machine, the first time an incident cannot be fully explained by looking at logs, or the first time a large customer's traffic pattern breaks the system for everyone else. These are different problems with the same shape — a system whose internal complexity has outgrown the mental model of the people maintaining it. Our job is usually less about new infrastructure and more about restoring that model.
Principals
R. Okafor
Reliability engineering
Previously staff SRE at two consumer platforms in the 30–80M MAU range. Writes the runbooks we leave behind. Maintainer of a widely-used open-source load-generation library.
J. Vasquez
Systems performance
Profiler-first engineer. Specializes in the gap between what an application thinks it's doing and what the kernel is actually doing about it. Occasional lecturer in graduate systems coursework.
S. Leclair
Search & retrieval
Information retrieval background, two decades of relevance tuning across marketplace, media, and legal verticals. Leads our search-shaped engagements.
D. Park
Platform architecture
Multi-tenant isolation, API design, and the politics of internal platforms. Joined the practice in 2021 after a decade leading infrastructure at a B2B SaaS platform.
A. Reinholt
Observability
Telemetry, cardinality, and the art of removing 80% of a metrics bill without removing any signal. Reviews every engagement's instrumentation plan.
What we decline
We do not accept equity. We do not take engagements under three months; scaling problems do not respond well to sprints. We do not work on greenfield systems — we are not the right team for your first prototype. And we do not take engagements where the primary stakeholder is outside the engineering organization; our work requires a senior engineering partner with the authority to make architectural decisions.
Where we work from
The practice is distributed across North America and Western Europe. We travel for kickoffs and cutovers. Beyond that, engagements run remotely with shared on-call shadowing during your business hours.