Staff Backend Engineer

About the Company A venture-backed data and AI company in private-market intelligence. They map a large and badly fragmented commercial landscape, tens of millions of organisations and hundreds of millions of the people inside them, and turn it into something you can search, compare and act on. The raw material arrives from hundreds of unstructured sources and has to come out the other side as a structured index a professional will bet a decision on. Under a hundred people, engineering-led, hybrid from a New York office. This role sits on the team that owns the core intelligence layer: the engine that decides what surfaces, in what order, for whom. They have shipped an agentic layer over that engine and it has become the part of the product customers came for. That is why they are hiring an architect for it rather than another contributor. The Role You would own the ranking, not tune it. Ingestion from hundreds of fragmented sources, a retrieval and reranking layer, personalisation with almost no clean training signal, and an agentic layer on top that is doing real work for real customers today. All four sit inside one system and nobody above you owns it. You would report into the engineering lead for this product area and hold the architecture together across the adjacent product, data and infrastructure teams. The stack: Python and Go on Google Cloud, Postgres and Elasticsearch, dbt over the warehouse, a hosted inference provider with an agent orchestration framework on top, TypeScript and React at the front. You own the retrieval, ranking, pipeline and orchestration parts of it and touch the rest. What is actually unsolved Relevance here is personal and there is no ground truth. Two people type the same query and want different results, and at the top of the funnel there is no click log deep enough to learn from. Where personalisation gets its signal is genuinely open.The system is non-deterministic. Deciding whether a prompt change, a model swap or a retrieval change was a real lift, and in which specific part of the product, is unfinished work, and doing it over a corpus this size is expensive on both cost and latency.The interaction patterns for agentic products are not settled anywhere. Long-running work, async results, and what a person sees while a system is still thinking. There is no playbook to copy and they are writing theirs. Responsibilities Own the architecture of the sourcing engine end to end, from ingestion through retrieval, ranking, personalisation and model orchestration, and set the patterns that let it power repeatable workflows across the rest of the product.Define the retrieval and reranking frameworks that hold result quality as the corpus, the query volume and the number of sources all grow.Extend the evaluation pipeline so a model, prompt or retrieval change can be shown to be a lift in a named part of the product rather than only in aggregate.Work with product and design on agentic interaction patterns: async work, long-running tasks, structured output, and what a person sees while the system is thinking.Raise the bar on system design, written technical decisions and mentoring across the engineering team. Requirements You have designed and scaled a search, recommendation or ranking system in production, and you can say what broke first as it grew.You have built and scaled data-intensive systems on asynchronous processing or event-driven pipelines, and you still owned them once they were live.Seven or more years building distributed systems or large-scale backend architecture.You can say what problem you were solving and for which group of users, without being asked. Weighted as heavily here as the systems depth.Clear progression at high-growth, venture-backed companies.A computer science or related technical degree. Nice to Have A track record of technical leadership and mentoring.You have led or heavily contributed to LLM features running in production, with real users on them.You have worked on entity resolution, knowledge graphs or messy third-party data at scale. Culture Small teams that own a product surface end to end rather than a service inside someone else’s. Engineering-led, and the people already there are a mix of former founders and engineers who have built at scale. Two things worth knowing before you decide. Their stated priority order is reliability, performance, cost, then developer productivity, and it is worth reading literally: developer experience work gets funded last. And they are writing the patterns for agentic products rather than adopting somebody else’s, which means a fair amount of what you build in the first year gets replaced by what you learn building it. If you want somewhere the architecture is settled and the job is to extend it, this is not that. Equity Offered on top of base Company Under a hundred people Discipline Backend and distributed systems