About The Team
The AI Lab is a small, founder-led team operating like a zero-to-one startup inside Life360, building the company’s next chapter. Our mission is to transform Life360 from the app you open to find your family into an intelligent operating system families rely on daily. We design and build an AI-powered layer on top of our family graph, leveraging our core location technologies and global 100 million-user base to proactively surface what matters across location, schedules, and everyday life.
About the Job
Your role is owning how Life360's deterministic backend (location data, the family graph, behavioral signals, user insights) pairs with the probabilistic world of LLMs. You decide how those two sides talk to each other, and you make sure the seams don't show.
You own the inference pipeline end to end: model selection, serving infrastructure, evaluation loops, prompt and context strategy, and the cloud services that keep it all running in production. You make the tradeoffs between latency, cost, and quality. You build the eval harness that tells us whether the system is actually getting better, not just feeling different. If you do this job right, the product feels seamless to users, the data stays safe, and inference costs don't run away from us. You are a cloud engineer who is deeply fluent in how AI systems actually work in production.
What You’ll Do
- Architect the inference pipeline that turns Life360 into a context layer LLMs can reason over: soul files, family context, behavioral signals, document understanding.
- Choose models. Choose hosting. Decide what is self-hosted, API-based, fine-tuned, or prompt-engineered. Revisit those decisions as the landscape changes.
- Build the serving infrastructure. Latency budgets, batching, caching, fallbacks, graceful degradation. Make it run at scale before scale gets here.
- Build the eval loop. Know whether changes make the product better or worse, not just whether the demo works. This is the difference between a real product and a magic trick.
- Own the cost model. Track spend per user, per feature, per family. Find the levers (context compression, model routing, pre-computation) that keep unit economics working from a test group to millions.
- Define what gets persisted, what gets summarized, and what gets thrown away. The architecture decision of "what does our data look like to an LLM" lives with you.
- Set the bar for safety, privacy, and trust at the infrastructure layer. Sensitive categories filtered before they propagate. Encryption, access controls, audit trails built in, not bolted on.
What We’re Looking For
If there's one trait we need above all others, it's self-starter energy. As we've gotten bigger, we've been more open to roles with clearly defined schedules and processes. That doesn't work in a startup. We need people on founder hours, not office hours — with a huge desire to win. You'll be a cofounder of this initiative.
- You operate without a spec. You'd rather have a conversation, a sketch, and 24 hours than a 30-page PRD and a week.
- You ship every day. Not "make progress." Ship. Even rough. Even ugly. Momentum matters more than polish at this stage.
- You push for more. You won't wait to be told what to do. You demand excellence of yourself and the people around you.
- You hold strong opinions and change them when the evidence moves. You push back. You concede fast when wrong.
- You're a customer of what we build. Try it if you haven't.
The Bar
- Deep engineering experience, with significant time spent building and running AI systems in production at meaningful scale. Real users, real load, real on-call.
- Fluent in modern LLM serving (vLLM, TGI, SGLang, hosted APIs — whatever's right for the job). No religion about it.
- You think in evals. You've built harnesses. You know the difference between vibes-based iteration and real progress because you've felt the pain of not having one.
- You can build a cloud system from the ground up. Pick the services, wire them together, make them reliable, keep the bill in check.
- Strong opinions about cost. You've seen inference bills run out of control and have instincts for where the leaks come from.
- AI-native in your daily workflow. Hands-on with Claude Code, Cursor, or equivalent. You think natively in agentic workflows, prompt engineering, context window management, MCP / function calling.
- Comfortable deciding with incomplete information. The spec will be a moving target. You don't need it to settle before you start building.
- Strong written communication. You write specs, decision records, and playbooks that an agent (and a human) can act on precisely.
- Experience with agent systems, retrieval pipelines, or long-context personalization.
- Privacy architecture experience in multi-user systems where context is shared across people with different rights to see it.
- You've built something where the data itself was the moat — and you understand why that's different from building on public data.