Staff Backend Engineer – AI

About Stream

Stream (GetStream.io)’s SDKs and APIs enable devs to build activity feeds, chat, and voice/video, powering over a billion end users worldwide. Trusted by top brands like Strava, Nextdoor, Patreon, and eBay, we’re on a mission to make real-time communication and social experiences seamless for developers and their users.

We’re also the makers of Vision Agents, the open-source Python framework for building low-latency voice and video AI agents: over 8k stars on GitHub, 35+ model integrations, and sub-500ms latency on our global edge network.

Role Overview

We’re seeking a Staff AI Engineer to own model development on our AI team. You’ll build, fine-tune, evaluate, and ship the models that run inside Stream’s products, end-to-end, from dataset design through to production. As the technical owner, you will drive key decisions independently, shipping models that directly impact systems serving over a billion users.

Location: Office in Amsterdam or Remote (Europe). Occasional travel for in-person collaboration and meetups is encouraged.

Hybrid policy: applicants based in the Netherlands or relocating here, are expected to work in the office in Amsterdam 3 times per week. Exemptions apply to specific cases.

What will you work on

• Own the development, fine-tuning, and evaluation of in-house AI models from dataset design through to production deployment.
• Run supervised fine-tuning and post-training experiments, establishing the benchmarks and evaluation harnesses that tell us whether a model is actually good enough to ship.
• Build and maintain the data pipelines that feed model training, keeping data quality, labelling, and reproducibility to a high standard.
• Take models to production on Stream’s serving stack, tuning for latency, cost, and reliability at high volume.
• Set the technical direction for an intentionally undefined problem space, deciding what to build, what to test, and what to abandon.
• Work across the wider engineering organisation, interfacing with Go-based API teams and infrastructure to get models integrated into the product.
• Contribute to the open-source ecosystem where relevant, and share work publicly through code, writing, or community engagement.
• Raise the bar on engineering standards across the AI team through code review, mentorship, and pragmatic best practices.

About You

• 5+ years of production-level Python engineering experience, with code you have shipped and maintained rather than only prototyped.
• Hands-on machine learning experience, specifically with supervised fine-tuning and post-training of models.
• Familiarity with the modern fine-tuning and serving toolchain, e.g. Unsloth, Fireworks, Baseten, or equivalents.
• Cloud experience with at least one major provider (GCP or AWS), including infrastructure-as-code with Terraform.
• Experience running ML-based products in production: not just training models, but owning them through deployment, monitoring, retraining, and iteration against real usage.
• Experience designing and operating data pipelines for training and evaluation; a data engineering background is a strong route into this role.
• Demonstrated ownership: a track record of picking up ambiguous problems and driving them to a result without waiting for direction.
• Strong communication skills and comfort working in a small, distributed, fast-moving team.

Preferred

• A visible open-source footprint: libraries you have authored or maintained, meaningful GitHub activity, or contributions to AI model repositories.
• Experience with Go (all of Stream’s APIs use Go, so it helps when interacting with other teams).
• Deep understanding of Python’s concurrency model and asyncio’s limitations in high-throughput systems.
• Experience with real-time or low-latency inference systems.
• Experience as an early engineer or founder, or otherwise operating at startup pace with an undefined roadmap.

Why You’ll Love This Role

• Genuine Ownership: You own model development end to end: approach, architecture, evaluation, and what ships. Not a slice of someone else’s roadmap.
• Your Work Reaches Real Scale Immediately: Models go into production across products serving over a billion end users. You’ll know quickly whether something worked, because real traffic will tell you.
• Small Team, Short Path: We’re small enough that there’s no layer between your decisions and their im…