At In Tandem , we build technology that helps families manage everyday routines and navigate life’s biggest transitions. Through our four brands—OurFamilyWizard, Cozi, FamilyWall, and Custody Navigator—we help families stay organized, communicate well, and foster healthy childhoods. We believe technology should strengthen relationships and make daily coordination less complicated.

Everything we create is designed to lighten the mental load, reduce conflict, and support families through big and small moments. If you want your work to make a real difference in the daily lives of parents and kids, In Tandem is the place where your impact will truly matter. As our AI Engineer , you'll keep the AI infrastructure our products and teams run on fast, efficient, and reliable, and you'll build with it.

You'll run and optimize our self-hosted inference stack on our own GPU hardware, build the internal platform our employees work through, and ship user-facing agents inside the apps. Your work spans OurFamilyWizard, Cozi, and FamilyWall, and the platforms that power how we build. This is a hands-on technical role at its core: you own the technical side of running our models on our own hardware.

But it's not siloed, and we don't want it to be. We're looking for someone who also wants to pick up app-layer work and ship product-facing features, and does both well. Run and optimize our self-hosted inference stack Run the inference serving layer on our own GPU hardware: choose and tune the serving stack (vLLM, SGLang, TensorRT-LLM) for high throughput and low latency.

Optimize aggressively: tensor parallelism, quantization (FP8, AWQ, GPTQ), KV-cache and prefix caching, continuous batching, speculative decoding, concurrency tuning. Serve multiple models and features off shared hardware: multi-LoRA, routing, and request scheduling that balances internal workloads against latency-sensitive product traffic. Keep our AI fast, efficient, and observable Make our AI workloads efficient: improve latency, throughput, and GPU utilization so we get the most out of what we run.

Build the visibility: instrument performance and usage across our AI surfaces so there's clear data on how everything is running. Surface the technical tradeoffs (performance, latency, efficiency) so the people making the calls have what they need to make them. Build AI features and proactive agents Ship the in-app agent layer that helps families coordinate: proactive nudges, smart suggestions, agents that summarize, draft, schedule, and act for busy parents.

Build the substrate underneath: tools, memory, orchestration, guardrails, and evaluation harnesses, integrated cleanly with production APIs alongside our architecture team. Work in nimble pairs with feature owners, standing up whatever's needed to test an idea, including a vibe-coded UI when that's the fastest path to a real customer. Ship rough, learn fast, harden what works.

Technical and hands-on with infrastructure: you like running real systems on real hardware and keeping them fast and reliable. A full-stack builder who wants the app layer too: you don't want to be boxed into infra. When a feature needs shipping, you want to pick it up and ship it, not just hand it off.

Performance-minded: you treat latency, throughput, and efficiency as things to engineer deliberately. Rapid-prototyping and AI-first, with modern tooling (Claude Code, agent SDKs) part of your craft. Motivated by work that matters.

Families rely on these products during real moments in their lives. 5+ years shipping production software, including meaningful applied AI or ML work. Demonstrated experience running and optimizing self-hosted LLMs on dedicated multi-GPU hardware: a serving stack (vLLM, SGLang, or TensorRT-LLM) and the optimization that comes with it (tensor parallelism, quantization, batching, KV cache).

A track record of optimizing inference performance and efficiency (latency, throughput, GPU utilization).