A high-growth AI-native startup is hiring an AI Engineer in New York to design, build and ship LLM-powered product features and agentic workflows into production.This is a build role focused on the applied AI stack: prompts, retrieval, tools, agents, evals and the glue that connects them to a real product. You’ll work close to product and real users, not inside a research lab.Role: AI EngineerSalary: $180,000 - $300,000 plus equity
Location: New York, 5 days per week in officeWhat you’ll be doing:Design and ship production LLM features and agents using Python and TypeScriptBuild the applied AI toolkit end to end: tool and function calling, RAG, vector search, structured output, prompt management, context handlingWork directly with foundation model APIs (Anthropic, OpenAI, Google) and open-source models where it makes senseBuild evaluation pipelines (Braintrust, LangSmith or custom) so quality is measurable rather than anecdotalIntegrate agents into existing product surfaces built in React and Next.jsOwn cost, latency and reliability of the AI layer as first-class concernsUse AI tooling (Claude, Cursor and similar) across your workflow to compress the build cycleWhat you’ll need:Strong Python and TypeScript, with real production experience shipping LLM-powered featuresPractical grasp of the applied AI stack: function calling, RAG, vector databases, structured output, prompt engineering, context managementComfort with at least one agent framework or SDK, with a view on why the abstractions existExperience building evals and guardrails for LLM outputSolid API, integration and backend engineering fundamentalsFluent in modern AI engineering tools and able to use them to raise outputHappy working 5 days a week in-person with a small, high-intensity teamNice to have:Prior work on agentic systems in production (not just prototypes)Experience with multimodal models (vision, audio, video)Exposure to fine-tuning or post-training of foundation modelsThis role would suit an AI engineer who wants to ship real LLM product into users’ hands, not write another whitepaper.