Research Engineer, Post-Training

ABOUT MACROSCOPE Macroscope is building the infrastructure for the next generation of software development, where engineers direct fleets of agents, every change is automatically reviewed and verified, and organizations have a clear understanding of how their products and codebases are evolving. Our mission is to give engineers more leverage and leaders more clarity, so teams can build better software, faster. Macroscope is founded by former entrepreneurs who have started and sold multiple companies, and operated as product/engineering executives at public tech companies. We're fortunate to be supported by the best VC firms and angels in the business, including Lightspeed Venture Partners, Thrive Capital, Google Ventures, and Adverb. ABOUT THE ROLE We're looking for an Research Engineer to improve the models and ML systems behind Macroscope's core AI capabilities. You'll work closely with our founders and engineering team to build evaluation datasets, run experiments, train and fine-tune models, and determine what meaningfully improves model performance. You'll have ownership across the ML lifecycle, including model evaluation, reinforcement learning, and bringing new approaches into production. You'll also explore emerging research and training techniques that can push what's possible with AI for software engineering. Our technology stack: TypeScript/React (front end), Golang (backend), Temporal, Google Cloud (GCP), Postgres, Terraform, and custom-built AST "code walkers" in Golang, TypeScript, Swift, Python, and Rust. QUALIFICATIONS - 3+ years of experience in applied machine learning, AI, or related engineering roles. - Experience building, training, fine-tuning, or evaluating modern ML models in production or research environments. - Experience with reinforcement learning for LLMs, including RLHF, RLAIF, GRPO, PPO, DPO, or similar techniques. - Strong experience with dataset creation, curation, benchmarks, labeling strategies, and evaluation methodologies. - Experience designing rigorous experiments and using results to improve model performance. - Familiarity with LLMs, reasoning models, and the evolving open-source model ecosystem. - Strong software engineering skills and experience building reliable ML pipelines and tooling. - Comfortable working in a fast-paced, high-agency startup environment. - Golang experience is a plus, but not required. - Bonus: Experience with distributed training, preference optimization, synthetic data generation, evaluation frameworks, GCP, Temporal, or internal ML tooling. The base pay for the Applied ML Engineer position ranges from $180,000 to $280,000 per year. Pay is based on several factors and may vary depending on job-related knowledge, skills, and experience. ABOUT YOU - You are extremely high agency. We are a small startup and we intend to keep an extremely flat organizational structure for as long as possible. Instead of relying on people managers, product managers and heavy processes, we rely on exceptionally talented individuals with high agency to be self-motivated towards contributing to our mission. - You want to work at an early stage startup. The default state of any startup is failure. The only way to overcome the daunting odds of making a startup venture successful is for a densely packed group of insanely hard working and talented people to work together to build something useful to and loved by customers. If you're not willing to work extremely hard on something high risk, this startup isn't for you. - You act like an owner. You put immense care and craft into what you build because you take responsibility for all parts of the product. You don't walk past broken windows. - You care about what we're building. Life's too short to work on something you're not passionate about. We are a small group of ambitious people who want to build something insanely great that we want to use, and that we think every company will want to use. If our mission and product don't resonate with you, we understand and would encourage you to find something that does.