Machine Learning Engineer

About Manifest

Manifest OS is an AI-native company on a mission to replace the billable hour and make legal services more accessible for American businesses and consumers. We power the next generation of AI-native law firms with one unified global brand, a proprietary technology platform, and a centralized back office — enabling lawyers to eliminate the administrative burden and focus on delivering exceptional outcomes for their clients. Backed by leading venture investors, Manifest O.S. is scaling rapidly.

About the team

Our ML team is a tight-knit group who are pushing the boundaries of what's possible with open-weight models. We move fast, experiment fearlessly, and measure everything that matters. The team collaborates closely with product and engineering to ensure our models don't just perform well on benchmarks — they make real users more productive and happy.

Ideal experience

You have 5-10+ years of hands-on experience with machine learning, with at least 2 years focused specifically on training or fine-tuning large language models. You're comfortable working with PyTorch and have experience with transformer architectures, understanding both the theory and practical considerations of training these models at scale. You've worked with modern post-training techniques like supervised fine-tuning, DPO, or RLHF, and you understand when and why to apply each approach.

What sets you apart is your ability to design clean experiments and interpret results honestly. You've built evaluation pipelines that actually matter, and you're not afraid to kill a promising experiment when the data doesn't support it. You have experience with dataset construction and curation, understanding how data quality impacts model performance in subtle but important ways.

What You'll Own

We're looking for a Machine Learning Engineer who gets excited about turning promising research into production-ready models that users love. You'll be at the center of our model development process, running experiments that matter and making data-driven decisions about what actually improves our product. This isn't just about making loss curves go down — it's about understanding what makes models useful in the real world and building the infrastructure to get us there consistently.

You'll own the full cycle from dataset curation to post-training optimization, working with cutting-edge techniques like supervised fine-tuning, DPO, RLHF, and distillation. The ideal candidate combines solid engineering fundamentals with research curiosity, always asking not just 'does this work?' but 'does this make our users' lives better?' You'll have the freedom to explore new approaches while maintaining the rigor to ship improvements that move the needle on product metrics.

This for you if

You're someone who thrives in the intersection between research and product. You love diving deep into the latest papers but you're equally excited about seeing your work impact real users. You have strong opinions about experimental design but hold them lightly when the data suggests otherwise. You're comfortable with ambiguity and enjoy figuring out the right metrics to optimize for, not just the easiest ones to measure.

You work well in a collaborative environment and communicate technical concepts clearly to both technical and non-technical stakeholders. You're self-directed but know when to ask for help, and you're comfortable with the iterative nature of research — most experiments don't work, and that's okay.

Not for you if

This role isn't ideal if you prefer working on well-defined problems with clear solutions, or if you're primarily interested in theoretical research without concern for practical applications. If you're someone who gets frustrated when experiments fail or when product requirements change based on user feedback, this fast-paced environment might not be the right fit. We need someone who cares about shipping improvements, not just publishing papers or achieving perfect benchmark scores.