Senior AI Engineer, Core Infrastructure

San Francisco or New York · 5 days in office - relocation required

$180K–$250K base + 0.3–0.8% equity

Visa sponsorship, transfers, and relocation available

This is an early-stage vertical AI company creating the agentic AI infrastructure layer for institutional real estate development.

The product automates complex development workflows across digital infrastructure, energy, industrial, and large-scale physical projects — the kind of work that usually moves through fragmented spreadsheets, documents, financial models, geospatial data, consultants, approvals, and manual coordination.

The company is already working with Tishman Speyer, Starwood, Meta, and Entropy. These are not lightweight pilots. They are some of the largest institutional players in real estate, infrastructure, and physical asset development.

The team has raised $8M and is backed by top-tier investors and senior operators, including C-suite leaders from Blackstone and OpenAI.

The engineering team is currently 4 people. A senior engineer joining now will shape the core architecture before it hardens.

You will own the core infrastructure behind long-running AI agents.

This is a backend and infrastructure role for someone who wants to design the systems that make agents reliable in production: orchestration, state, retries, fallbacks, resumability, retrieval, evals, observability, and deployment.

The work sits close to the product surface, but the center of gravity is systems engineering. You will not be polishing chatbot flows. You will be deciding how agents execute multi-step workflows over messy enterprise data without silently failing, losing state, hallucinating context, or breaking under real customer usage.

You will work directly with the technical leadership and the rest of a small engineering team. There is no thick management layer, no narrow ticket queue, and no mature platform to inherit.

Real estate development is a high-stakes, multi-party workflow where bad information creates expensive downstream errors.

An agent may need to reason across site data, zoning constraints, financial assumptions, construction documents, geospatial context, emails, internal knowledge, and customer-specific processes. The hard part is not calling a model. The hard part is making the system reliable when the workflow is long-running, partially ambiguous, and tied to decisions worth millions of dollars.

That creates pressure on:

• Agent runtime design: tool orchestration, workflow state, retries, fallbacks, cancellation, resumability, and idempotency.
• Context architecture: retrieval over financial, geospatial, document, and customer-specific data.
• Production reliability: queues, traces, alerting, replayability, observability, and failure recovery.
• Evaluation: catching regressions in agent behavior before customers do.
• Data infrastructure: Postgres, columnar stores, queue systems, cloud services, and pipelines that stay understandable as complexity grows.

The easy version is a demo that works once.

The useful version is a system that can be trusted across repeated, high-value customer workflows.

• Core agent runtime: design and evolve orchestration, workflow state, retries, fallbacks, resumability, and failure handling for long-running agents.
• Tool execution infrastructure: create the backend systems that let agents call tools safely, recover from partial failures, and preserve useful state across multi-step workflows.
• Retrieval and context systems: design how agents retrieve, rank, and use context from documents, geospatial data, financial models, and structured internal systems.
• Production reliability: own monitoring, tracing, alerting, CI/CD, deployment practices, and operational visibility for AI systems used by real customers.
• Data and systems architecture: work with Postgres, columnar databases, Kafka-like queues, AWS services, and distributed systems patterns to keep the platform reliable as usage grows.
• Eval infrastructure: design trace replay, regression detection, quality measurement, and test harnesses that make agent behavior easier to inspect and improve.
• Architecture decisions: make early technical calls on language, infrastructure, interfaces, and system boundaries that the company may live with for years.
• Cross-stack execution: work across backend, infra, product constraints, and customer workflows when the highest-leverage solution does not fit neatly into one layer.

You are likely a strong fit if you have:

• 5–10 years of backend or infrastructure engineering experience, with meaningful ownership of production systems.
• 3–4 years in early-stage environments, ideally Seed to Series B, where you owned broad technical scope without heavy process.
• Strong backend fundamentals across distributed systems, databases, queues, observability, CI/CD, and cloud architecture.
• Production experience with Python, Go, or Rust, and the judgment to explain when each language is the wrong tool.
• Experience with AWS, especially services such as ECS, Lambda, S3, or similar production infrastructure.
• Experience with Kafka or comparable queue/event systems, including failure modes, retries, ordering, backpressure, and operational trade-offs.
• Strong opinions about system design, but enough discipline to change your mind when production evidence disagrees.
• Comfort with ambiguous requirements, customer-specific workflows, and technical decisions where there is no established playbook.
• Interest in AI systems where the challenge is not prompt writing, but making agents dependable inside real business processes.

Top candidates will be able to explain how they would design an agent runtime that fails visibly, recovers cleanly, and produces enough traces to debug after the fact.

• Languages: Python, Go, Rust
• Backend/data: PostgreSQL, columnar databases, Kafka or similar queues
• Cloud: AWS, ECS, Lambda, S3, Cloudflare
• AI/application layer: LangGraph, retrieval systems, agent orchestration
• Data infrastructure: Databricks and related analytical systems
• Platform: CI/CD, observability, monitoring, tracing, alerting

The stack is still early. Good judgment matters more than exact tool overlap.

The company is at the stage where the core technical foundation is still being decided.

The customer demand is real, the workflows are complex, and the engineering team is small enough that one senior hire can materially change the architecture, development velocity, and product reliability.

This is the highest-leverage moment to join: before the agent runtime, data architecture, eval infrastructure, and operational practices become fixed.

For someone choosing between a large AI lab, Big Tech L6/L7 scope, and an early-stage company, the trade-off is clear: less infrastructure already exists, but far more of the company’s technical surface is still available to own.

• You want a narrow model integration role.
• You prefer mature infrastructure with clear ownership boundaries.
• You want a PM to fully define requirements before you start.
• You are uncomfortable being accountable for production reliability.
• You do not want to reason about queues, state, databases, observability, and deployment.
• You are looking for remote-first work.
• You want AI work that is mostly research, prompting, or frontend product iteration.

• Base salary: $180K–$250K
• Equity: 0.3–0.8%
• Location: San Francisco or New York
• Office policy: 5 days in office
• Visa: sponsorship, transfers, and relocation available
• Team size: 11 people total, including a 4-person engineering team
• Funding: $8M raised
• Stage: early-stage, founded in 2024

The balance between cash compensation and equity can be tailored to your preference.

Aurora helps exceptional engineers find the right role at some of the most ambitious startups worldwide.