Benefits
- Equity
Skills
Nice to have
Full job description
- Designs, develops and manages the lifecycle of a product or group of products from concept to launch to end of life. Translates market opportunities and customer demand into viable products and services that differentiate Equinix in the market
- Sets the vision and strategy for their product ensuring it is competitively positioned and customer-centric. Manages the product roadmap including features, upgrades and maintenance of the product or product line
- Works cross functionally with user experience, engineering, operations, solution architects, marketing and others to design, build and launch new products and/or product features
- Define how Equinix chooses models for each application. Criteria include task quality on our own evals, cost per task, latency, context length, tool-use ability, hardware footprint, license terms and data-handling requirements
- Maintain a recommended-model catalogue across open-weight and proprietary models, with clear guidance on when each fits
- Track new model releases, benchmark them against Equinix workloads, and decide when an upgrade is worth the switching cost
- Set the playbook for adapting models, from prompt and context engineering and retrieval to supervised fine-tuning, parameter-efficient methods (such as LoRA), preference optimization, reinforcement learning from feedback, distillation and quantization
- Help teams choose the lightest technique that meets the bar. Prompting and retrieval come first, and training is used where it clearly pays off
- Define the data strategy behind tuning, including curation, labeling, synthetic data generation, deduplication, and protecting against leakage between training and evaluation data
- Own the roadmap for a shared model customization platform covering data preparation, training, evaluation, registry, deployment and lifecycle management
- Partner with infrastructure and engineering teams on GPU capacity, training and serving stacks, inference optimization, and cost controls
- Make tuned models reproducible and supportable, with versioned data, configs and checkpoints, clear ownership, and rollback
- Work with business and domain teams to scope use cases, pick models, design tuning experiments and launch to production
- Turn early wins into repeatable patterns, templates and services that the next team can reuse
- Run model launch reviews covering quality, safety, cost, performance and operational readiness
- Define the evaluations every model must pass before launch: task-specific quality, regression against the base model, safety, robustness and cost and latency targets
- Partner with the Agent Intelligence team so tuned models are measured in production and feedback flows back into better data and better models
- Report the impact of customization in terms leaders care about, such as quality gained, cost saved, latency reduced and risk avoided
- Set standards for license review, model provenance, data rights, model documentation, red-teaming and approval to deploy
- Make sure confidential and customer data used in tuning is handled within Equinix’s security, privacy and regulatory requirements
- Be Equinix’s go-to product leader on model selection and customization, and influence teams you don’t manage
- Write the strategy, specs and executive briefs that align engineering, data science, infrastructure, security and business leaders
- In your first year, you will
- Publish Equinix’s model selection framework and recommended-model catalogue, backed by benchmarks on our own workloads
- Launch a shared customization platform that takes a team from data to a deployed, evaluated, tuned model
- Ship tuned open-weight models into production with at least two vertical teams, with measured gains over the baseline
- Establish a governance process for licensing, data rights and approval to deploy that teams can move through quickly
- This is not a research scientist role, and it’s not a pure prompt-engineering role. You will own the product: the framework, the platform and the launches. You will be accountable for tuned models that are measurably better, cheaper or faster in production
- What success looks like:
- Every team knows which model to start with and why, and model choices are backed by Equinix’s own benchmarks
- Teams reach for fine-tuning only when it pays off, and when they do, the platform makes it fast and repeatable
- Tuned models in production beat their baselines on quality, cost or latency, and the gains are measured and reported
- Model launches pass a clear, quick governance process with no surprises on licensing, data rights or safety
- Equinix builds lasting expertise in running and improving its own models
- Why this role matters:
- Choosing and adapting models well is becoming a core capability for any company building with AI. Open-weight models let Equinix put the right model on each problem, control cost and data, and build behavior that general-purpose models can’t match. You will make that capability real, repeatable and trusted across the company
- Hands-on experience customizing and shipping open-weight models in production, such as fine-tuning, parameter-efficient tuning, preference optimization, distillation or quantization
- Working knowledge of the model training and serving stack (for example PyTorch, Hugging Face, and inference servers such as vLLM or SGLang) and of GPU cost and capacity trade-offs
- Comfort reading and writing Python, running experiments, and reviewing training and evaluation results yourself
- Strong evaluation skills, including building task-specific evals, avoiding train/test leakage, and telling a real improvement from noise
- 7+ years across product management, engineering or applied ML, including 3+ years as a product manager on AI/ML, data or infrastructure platforms
- Ability to lead without authority across engineering, science, infrastructure and business teams, and to explain trade-offs clearly to senior leaders
- Deep practical knowledge of prompt and context engineering and retrieval-augmented generation, and sound judgment about when to use them instead of training
- Experience selecting models for real applications, weighing quality, cost, latency, hardware and licensing
- Experience with tool-use or agentic fine-tuning, or with function-calling models
- Experience in regulated industries such as financial services
- Experience with reinforcement learning from human or AI feedback, or with reward modeling
- Experience with synthetic data generation and data curation at scale
- Experience with open-weight model licensing, model governance or responsible AI programs
- Experience running models in private, hybrid or on-premises environments, or with data residency requirements
- A degree in computer science, engineering, machine learning or another quantitative field, or equivalent experience. Publications or open-source contributions are a plus
Opens Equinix’s own application page in a new tab.