Staff Software Engineer

Staff Engineers | Python | $200,000 to $300,000 + equity | New York Series A AI TeamIf you are a serious engineer evaluating AI companies right now, you are probably asking one question: what happens to this product when Anthropic or OpenAI releases the next model? Here is the honest answer for this one.The productThe firms that finance the built world spend their days inside documents. Lease portfolios. Data rooms assembled for deals worth hundreds of millions. IC papers that have to be right, cited, and auditable. For decades they have done this work with Excel, PDFs, and analysts working late.This platform turns all of it into structured intelligence that investment teams actually trust. A PE firm with £1.5bn AUM went from days to 30 minutes on an IC paper. 1,700 leases across 100 assets, from two weeks to half a day.Enterprise clients across the US, Europe and APAC. ARR up 6x last year. Series B next year.Why a new model release makes this stronger, not redundantThe platform runs multiple foundation models in parallel with automatic failover. A better Claude or GPT is a stronger component inside infrastructure you helped build.What you are building is the defensible part: a proprietary embedding model that outperforms OpenAI, Google and Nvidia on domain-specific document retrieval; a novel transformer architecture published on arXiv; evaluation infrastructure holding 98.5% accuracy against real client documents.The clients deepen the moat every month: investment criteria embedded, corrections fed back, data rooms integrated. No model release replicates that.The team you would be joiningA CPO co-founder with 17 years applying data and analytics to institutional real estate, who knew this problem from the inside before building the product.A CEO co-founder with a Stanford PhD and a background as CPTO at a Series A fintech, one of very few female technical founders at Series A.A CTO who was at Wise pre-IPO, built two companies from zero, is the named technical voice in a published Google Cloud case study, and invests as a first cheque in applied AI and dev tooling.A Principal Research Scientist with two arXiv publications, one introducing a novel transformer architecture built inside this team.Alumni of Babylon Health, GRAIL, Hadean and Tractable. Non-linear paths, elite output.Your weekMonday: experiments from the weekend surface something worth pursuing. Hypothesis shipped before lunch using Claude Code.Wednesday: you are on a call with a major asset manager in New York, spotting a problem nobody has named yet.Thursday: spec written, building started.Friday: the client says this is exactly what we needed.You own product areas end to end from discovery through to customer adoption. The architectural decisions that define this category are still ahead of you. Feedback loop is days, not quarters.What this requiresStrong Python throughout, full-stack across backend services, APIs and frontend.The stack is GCP-native: Vertex AI, event-driven architecture, multi-model orchestration across Claude and Gemini.You have shipped agentic workflows in production, not prototyped them.You understand evaluation-driven development, build evals before you ship, and can tell when a model output is wrong without a test suite to tell you.Experience with retrieval systems, document processing pipelines or LLM orchestration at enterprise scale.You have worked in B2B where your code directly affected a customer's business outcome and you can say what that outcome was.Claude Code is your primary development tool, not a supplement.The packageUp to $300,000 base + meaningful Series A equityUnlimited leave, enhanced parental leave, and a wellbeing support package that goes well beyond the standard.Hybrid, 2-3 days in the New York office.I have placed two engineers here in the last month. They still need five more due to client demand. Three stage process, turned around in one to two weeks.Offers made within 24 to 48 hours of final stage.Apply via LinkedIn or email reuben@techmunity.io