Staff Backend Engineer

About the CompanyOur client builds real-time voice AI infrastructure for enterprises in regulated environments, at production volume, for customers who cannot tolerate it failing. Enterprise security and healthcare compliance are already in place. Seed round closed, led by a well-known fund, with operator angels from several recognisable technology companies on the cap table.The team is still in single digits. Senior-heavy, flat, no engineering management layer, everyone reports to the founder, who holds both the CEO and CTO seat. They were early to their market and have kept the lead through head-to-head technical evaluations. Enterprise demand now runs ahead of what the current architecture can carry, which is why this role exists.You would lead that work, alongside owning the core backend services that keep the real-time platform standing while it scales. This is their top-priority hire and they are making exactly one. You report directly to the founder, so there is no layer between you and the person making the technical calls.What is actually unsolvedWhat the storage layout should be when one record is an entire conversation, with audio, transcript, traces, model calls and outcomes attached, and queries run from "show me this one" to aggregating across millions.Where the boundary sits between the new store and the Postgres and ClickHouse already carrying production.What degrades first at target concurrency, and whether it degrades gracefully or simply falls over. Nobody has proven this yet.How you make a non-deterministic pipeline debuggable, so "why did this one go wrong?" has an answer that is not an hour of log reading.ResponsibilitiesOwn the data layer end to end. What a record is, how it sits on disk, and which access patterns get to be fast. Owning it means the pager, not only the design.Take throughput up by close to an order of magnitude against a tight availability target. Nobody currently knows what breaks first, and finding out before a customer does is the job.Make production monitoring trustworthy at volume, so events and outcomes never quietly go missing. Silent data loss is worse than an outage, because nobody notices it.Take the question of why one request went wrong from an hour of log reading to a thirty-second answer, on a pipeline where the same input can produce a different output.Ship the smallest working version in week one. Designs get refined against production rather than in a document.RequirementsYou have built or core-contributed to a database, a queue or a large-scale data system. You can name the subsystem, the decision you made inside it, and what that decision cost you. Operating one at scale is a different thing and the interview finds the difference quickly.You have taken something from working to working under load. Gigabytes to terabytes a day, high cardinality, tight service levels. You found the bottleneck yourself, you can still quote the number before and after, and you know which of your fixes was the one that actually mattered.Internals-level with at least one of Postgres, Redis, Kafka or ClickHouse. Internals means you know how it behaves when it is unhappy, not that you have configured it.AI-native, genuinely. Cursor, Claude Code or Codex as daily drivers, with a view on which model you reach for and where each one lets you down. Enthusiasm with no criticism reads as shallow use.CultureSmall team, senior throughout, flat. No engineering management layer and no ladder. Decisions get made in the room by the people who will implement them, and nobody has a narrow lane.Deploys go out several times a day and fixes often ship the same day a customer raises them. There is no QA function between you and the customer, and whoever writes it carries the pager. Prototypes take days rather than months, and long whiteboard cycles do not survive here.