Applied ML Engineer
Macroscope2 months ago
San Francisco, CA, USAMid Level
Base Salary
$170k - $280k/yr
Responsibilities
- Build and maintain high-quality evaluation datasets and benchmarks.
- Design and run experiments to assess model performance and determine why models perform as they do.
- Train and fine-tune models, including reinforcement learning where appropriate.
- Analyze experimental results and use findings to drive model and product improvements.
- Participate in research on advanced model-development and reinforcement-learning techniques.
- Partner with product and backend engineering teams to integrate new models into production.
- Help shape the future of AI-powered software engineering at Macroscope.
Requirements
- At least 3 years of experience in applied machine learning, AI, or related engineering roles.
- Experience building, training, fine-tuning, or evaluating modern machine learning models in production or research environments.
- Experience with reinforcement learning or LLM reinforcement learning, including techniques such as RLHF, RLAIF, GRPO, PPO, or DPO.
- Strong experience creating, curating, and evaluating datasets, including benchmarks, labeling strategies, and evaluation methodologies.
- Experience designing rigorous experiments, analyzing results, and using data to improve models.
- Familiarity with LLMs, reasoning models, and the open-source model ecosystem.
- Strong software engineering skills and experience building reliable ML pipelines and tooling.
- Experience with Golang is a plus but is not required.
- Bonus experience includes distributed training, preference optimization, synthetic data generation, evaluation frameworks, GCP infrastructure, Temporal, or internal ML tooling.
Tech Stack
Categories
About Macroscope
Macroscope superpowers software teams so that everyone has the advantage. Leaders get an accurate birds eye view of how the codebase and product is evolving, on-demand without ever having to interrupt engineers. Engineers save time by getting direct access to all the information about the codebase and automate code reviews that find correctness issues before they get to production. And we power all of this from a single source of truth - the codebase itself.