Software Development · Austin, TX

On-site Full-time

Software Engineer, Backend - Distributed Compute

Base Power · Austin, TX · Posted Nov 8, 2026

This posting comes from Base Power. Confirm the details on their site before applying.

Overview

Base's Distributed Compute effort turns our residential power network into a distributed GPU inference cloud. The control plane that makes this possible is responsible for fleet management, network configuration, resource scheduling and telemetry. We're hiring a Software Engineer to own and harden the backend serving path: from our OpenAI-compatible API, through the relay layer, to compute nodes in homes.

Responsibilities

  • Build the backbone that connects a customer’s AI request to a GPU sitting in a real home, and make it fast and dependable at scale.
  • Take Distributed Compute from a handful of pilot homes to thousands of nodes, and shape how the whole fleet runs.
  • Keep every request quick: measure the delay our network adds, set an ambitious target, and hold the line.
  • Ship updates without anyone noticing, with zero downtime.
  • Make onboarding effortless and secure, bringing 100+ new homes online every week.
  • Build the smart routing that sends each request to the right machine, and the testing to prove it holds up under load.
  • Design systems that keep working when homes go offline, connections drop, or things break.
  • Own what you build end to end: design it, instrument it, run it in production, and carry the pager for it.
  • Help create a new kind of computing that turns the power grid into the AI era’s infrastructure.

Requirements

  • Distributed systems: you design and ship correct concurrent systems, including long-lived bidirectional gRPC streams, session ownership, leases and heartbeats, and graceful degradation.
  • You reason clearly about failure modes, backpressure, and exactly-once vs. at-least-once semantics.
  • Kubernetes and cloud infrastructure: you operate production EKS services confidently, with a working knowledge of informers, CRDs, leases, garbage collection semantics, and IAM/STS auth flows.
  • You can debug the "the pod died and routers didn't notice for five minutes" class of problem.
  • Ownership and operational scrappiness: you thrive with ambiguity, instrument what you build, and unblock yourself.
  • You're comfortable being paged for a system you wrote, using the paging system you configured.

Preferred qualifications

  • Networking and edge connectivity: deep comfort with NAT traversal, outbound-session architectures, TLS/mTLS, and overlay networks
  • LLM inference serving: hands-on experience with vLLM or similar, including OpenAI-compatible APIs, streaming, KV cache behavior, quantization tradeoffs, and TTFT/throughput benchmarking.
  • You know the difference between "the model loads" and "the node serves economically."

Great fits start here.

Whether you're looking for your next role or your next team member, give people and possibilities a place to meet.