Founding AI Engineer

We are a pre-seed team, well funded by a top European fund and a strong group of angels and operators. We're building AI agents for GTM — agents that do the work of a revenue team end-to-end: researching accounts, drafting outreach, updating the CRM, triaging inbound, prepping calls, following up. GTM data is the messiest data in the enterprise. It lives across 10+ systems, it's stale, it's permissioned, and full of half-truths people typed into Salesforce on a Friday afternoon. Building agents on top of this without hallucinating and actually getting the job done at scale is the real engineering problem.

About the role: You'd be the first engineer. You set the bar for architecture, evals, and what great looks like. The hard part isn't the agent loop — it's the data and context layer underneath: pulling from Salesforce, HubSpot, Gong, Slack, Notion, Drive and a long tail of broken APIs, then turning that mess into the precise, fresh, permission-aware context an agent needs to take a real action. You'll build both the context layer and the agents on top. Most of your hardest engineering time will be on the data side: ingestion, modelling, freshness, retrieval quality, evals, latency, permissions.

First 90 days: Own the data and context pipeline end-to-end: connectors, ingestion, schema modelling, chunking, embedding, indexing, hybrid search, reranking, serving. Own the agent runtime on top: tool use, planning, multi-step execution, error recovery. Build the eval harness before shipping anything new. Retrieval evals (recall@k, MRR, faithfulness), agent evals (task success, tool-call correctness, hallucination rate), and a CI gate that blocks regressions. We don't ship on vibes. Make the spine decisions: vector store, embedding model, reranker, agent framework (or none), orchestration, observability. By month six you'll have hired the second engineer and shaped the team's culture around data quality, evals, latency, and product taste.

Required: 5+ years of production backend experience, with meaningful time on data-heavy systems — retrieval, search, recsys, streaming pipelines, or ML infra with real users and real SLAs. You think data first. Schema design, freshness SLAs, idempotency, dead-letter handling, schema drift, CDC — second nature. Bad data in, bad agents out, and you've felt that pain. Shipped a non-trivial RAG, semantic search, or agent system end to end. You can talk fluently about why you chunked the way you did, how you measured it, and what broke. Strong opinions on hybrid retrieval. You know when BM25 beats embeddings, when SQL beats both, and you've made those calls in production. Eval-obsessed. You've built (not just used) eval pipelines for retrieval and agent behaviour. Comfortable owning infra without a platform team. Latency-budget thinking is second nature.

(nice to have) Python or Go fluency. Strong signal: Agents in production, not just RAG. You know what tool-use looks like with 40 tools, and how to keep an agent from looping forever. Multi-tenant systems with row- or document-level access control. You know what happens when permissions desync from the index. Messy GTM stack integrations — Salesforce, HubSpot, Gong, Outreach, Apollo — and the scars to show for it. Data governance or security overlap: lineage, audit trails, PII handling, prompt injection defence.

Interview Process: Four stages, no filler. Founders call (60 min, remote). What you've built, what you'd own here, whether we'd want to spend years in the trenches together. Technical deep-dive 1 (90 min). Walk us through a data-heavy system you've built. Technical deep-dive 2 (180 min, in-person, London). You and the technical founder build something end-to-end together.

Offer: What we offer top-of-market base for our stage. First-engineer-grade equity. Four-year vest, one-year cliff. Hardware budget you don't have to justify. 25 days holiday + UK bank holidays.