Founding Developer Relations Engineer

Goaly is coming out of stealth, and the company needs a technical voice who can make post-training performance tangible for the people running it day to day. As the first Developer Relations hire, you will build the benchmarks, tooling, and documentation that practitioners rely on, then distribute those artifacts across the channels where ML engineers actually learn and share.

What you will do

• Build the signature technical artifacts, including a public post-training efficiency benchmark, reproducible before/after case studies (including doomed-run kills, silent throughput regressions, and dollars per eval point), and the open-source tools released around them.
• Write conversion-focused docs such as quickstarts, reference documentation, and example integrations for the stacks users run, including PyTorch distributed, Megatron, FSDP, TorchTitan, Ray, Slurm, Kubernetes on GPUs, Weights & Biases (W&B), and MLflow. The goal is to help an engineer generate a read-only Retroactive Kill Report on their own cluster without needing Goaly’s involvement.
• Create practitioner-ready technical content, including deep-dive engineering posts, benchmark write-ups, failure-mode teardowns (for example, “seven ways an RL run dies silently”), short demo videos, and live walkthroughs.
• Own distribution by running Goaly’s technical presence across X, LinkedIn, GitHub, Hacker News, and the ML-infra Discord and Slack communities, plus deliver talks at venues users attend such as PyTorch Conference, MLSys, Ray Summit, GTC, and NeurIPS and ICML workshops, along with Bay Area AI infra meetups.
• Act as a feedback loop: sit with design partners, reproduce their problems on Goaly’s stack, and deliver precise, prioritized engineering feedback. Convert repeated questions into docs, tools, or product changes.
• Build community around the problem, not the product, through practitioner conversations, office hours, and a newsletter or series focused on learning about post-training efficiency.
• Amplify the founders by turning the founding team’s Llama post-training infrastructure experience into content and talks, coordinating with the GTM team to strengthen both the design-partner pipeline and candidate pipeline.
• Measure and improve by tracking reach, repo adoption, the docs funnel, inbound design-partner requests, and hires sourced. Double down on what works and stop what does not.

How the work is split

• About half of the week is building: running real post-training workloads on Goaly’s stack, creating benchmark harnesses, building example integrations, and writing docs that take someone from zero to a first result in about thirty minutes.
• About half of the week is distribution and relationship building: engineering posts, talks, open-source releases, and one-on-one conversations that lead to design partners and hires.

Who you are

• 3+ years as a software or ML engineer, with the expectation that you still write code every week.
• A strong engineering background: you can stand up a distributed training job, read a profiler trace, and debug a CUDA OOM without needing help.
• Hands-on experience with LLM post-training and/or the infrastructure underneath it, including SFT, RL, LoRA, DPO, or the stack beneath them such as PyTorch distributed, Megatron, DeepSpeed, FSDP, Ray, Slurm, and GPU Kubernetes.
• A track record of explaining difficult topics clearly through blog posts, docs, conference talks, OSS READMEs, tutorials, or videos (links are expected).
• Distribution you can point to, such as an audience on X, LinkedIn, YouTube, a newsletter, or GitHub, plus talks given, communities where people know your name, or evidence of building reach from zero in ML infrastructure and post-training.
• Two-way communication: write content practitioners trust, and also listen to a frustrated infra engineer for an hour and return with the three engineering fixes that matter most.
• A taste for substance: you would rather publish one reproducible benchmark than multiple hot takes.
• Comfort being the public face of something early, even without a playbook, with messaging that shifts as the product evolves.

Technical stack you will work with

• PyTorch distributed, Megatron, FSDP, TorchTitan, Ray, Slurm, Kubernetes on GPUs, Weights & Biases (W&B), MLflow
• SFT, RL, LoRA, DPO, DeepSpeed
• CUDA OOM, W&B, Nsight, PyTorch Profiler

Location and logistics

• Palo Alto, CA (hybrid): 4+ days/week in office, with roughly 20% travel for conferences and customer visits.
• Visa sponsorship: H-1B and OPT/CPT support available, with immigration counsel.
• Perks: complimentary lunch, dinner, snacks, and drinks.

About the role, in practice

Goaly’s users are ML infra engineers and researchers who operate SFT, RL, and fine-tuning jobs on GPU clusters. They do not read marketing; they read benchmarks, repos, and documentation, and they rely on the engineer behind the artifacts. This role reports to the founders, works daily with engineering and the go-to-market team, and is explicitly founding-stage: you help shape the playbook while keeping output credible to people who train models.

How Goaly works

• Mission first: choose work for impact and take responsibility for outcomes.
• High agency: identify what is missing, form a plan, and move without waiting for perfect clarity.
• Speed with rigor: ship, measure, and iterate while protecting correctness, safety, and reliability.
• Flexible scope: cross team boundaries when it is the fastest path to the real problem.
• Low ego, high standards: give direct feedback and adjust when evidence changes.
• Continuous learning: the stack changes quickly, and you are willing to learn unfamiliar systems and domains.