ML Engineer, Post-Training and Evaluation

Reflection • San Francisco, CA

11h • Onsite

About The Position

Reflection’s mission is to build open superintelligence and make it accessible to all, developing open weight models for individuals, agents, enterprises, and even nation states. The team consists of AI researchers and company builders from DeepMind, OpenAI, Google Brain, Meta, Character.AI, and Anthropic. This role is for a core member of Reflection's Applied AI team, focusing on driving model fine-tuning and evaluations for enterprise customers. The Applied AI team adapts Reflection's open-weight models for specific customer domains, tasks, and constraints. As an ML Engineer, you will work hands-on with customer data, run fine-tuning workflows, build evaluation harnesses, and deploy adapted models to production. You will work directly with customers to understand what they need and with research teams to push what's possible.

Requirements

• Applied ML experience with hands-on fine-tuning of language models. You have prepared datasets, run training loops, evaluated results, and shipped a fine-tuned model. Familiarity with SFT, DPO, RLHF, or similar techniques.
• Understanding of evaluation methodology: how to design evals, interpret training graphs, and tell whether a model is actually better or just overfitting to the benchmark.
• Comfort with training infrastructure: GPUs, compute management, debugging common training failures. You don't need to be an infra engineer, but you should not be afraid of a stack trace from a training loop.
• Strong software engineering fundamentals (Python). You write clean, reproducible code. Experience with data pipelines and version control for datasets and experiments.
• 3+ years of engineering experience with meaningful exposure to applied ML or ML engineering (e.g., MLE, Applied Scientist, Data Scientist who shipped models to production, or ML-focused SWE).
• Demonstrated ability and interest to work in customer-facing environments, understanding user needs and translating domain requirements into training strategies.
• Self-starter with high agency and ownership, excelling in fast-paced startup environments where playbooks are still being written.

Responsibilities

• Fine-tune Reflection's open-weight models for customer-specific use cases: prepare datasets, configure training runs (SFT, preference optimization, reinforcement fine-tuning), and iterate based on evals.
• Build and maintain evaluation infrastructure: design eval suites, curate test sets, establish baselines, and measure whether fine-tuned models actually improve on the tasks customers care about.
• Prepare training data from raw customer inputs: inspect data quality, clean and format datasets, identify adversarial or noisy samples, and build reproducible data pipelines.
• Debug and diagnose training and inference issues: interpret loss curves, catch data quality problems, and identify when training dynamics indicate something is wrong.
• Support end-to-end deployments of fine-tuned models across hybrid environments (public cloud, VPC, and on-premises), helping ensure inference performance and reliability in production.
• Contribute to evolving playbooks, evaluation benchmarks, and best practices as part of a growing fine-tuning and evals practice.

Benefits

• Top-tier compensation: Salary and equity structured to recognize and retain the best talent globally.
• Comprehensive medical, dental, vision, life, and disability insurance.
• Fully paid parental leave for all new parents, including adoptive and surrogate journeys.
• Financial support for family planning.
• Paid time off when you need it
• Relocation support
• Lunch and dinner are provided daily.
• Regular off-sites and team celebrations.

What This Job Offers

• Job Type: Full-time
• Career Level: Mid Level
• Education Level: No Education Listed
• Number of Employees: 1-10 employees