Mid-Level 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

You'll join our team that bridges research and product, working closely with our engineering and product teams to ship models that actually improve user experience. We're pragmatic researchers who care as much about deployment metrics as we do about loss curves, and we value rigorous experimentation over flashy demos.

Ideal experience

You have 3-5 years of experience working with machine learning models, with significant hands-on experience training or fine-tuning transformer models using PyTorch. You understand modern post-training methods like supervised fine-tuning, DPO, and RLHF not just conceptually but practically — you've implemented them, debugged them, and seen what works and what doesn't in real scenarios.

You approach ML work like a scientist: you design clean experiments, control for variables that matter, and read results honestly rather than cherry-picking what supports your hypothesis. You're comfortable with the full experimental lifecycle from dataset creation through evaluation pipeline setup to analyzing results and deciding what to try next. You stay current with ML research but you're selective about what's worth implementing versus what's just interesting.

What You'll Own

We're looking for a mid-level ML engineer who will own the model training and fine-tuning pipeline that powers our core product features. This isn't just about getting models to work — it's about making them work well for real users. You'll run experiments on open-weight models, build the datasets that teach them what we need, and implement post-training techniques that turn raw capability into genuinely useful behavior.

This role sits at the intersection of research and product development. You'll need to stay current with the latest techniques in SFT, DPO, and RLHF, but your success will be measured by whether the models you train actually make our product better. You'll design experiments that answer real questions, interpret results honestly even when they're disappointing, and iterate quickly based on what you learn.

The impact is immediate and measurable. The models you train will serve millions of users, and you'll see directly how your work affects product metrics, user satisfaction, and business outcomes. This is a rare opportunity to do cutting-edge ML work while building something people actually use.

This for you if

You're the kind of person who gets excited about both the latest research paper and the production metrics dashboard. You have strong research instincts — you form hypotheses, design experiments to test them, and follow the data wherever it leads. But you also care deeply about whether your work actually makes the product better, not just whether it advances the state of the art.

You thrive in environments where you can work independently but collaborate closely with others. You're comfortable taking ownership of complex technical problems while also communicating your progress and findings clearly to both technical and non-technical stakeholders.

Not for you if

If you're primarily interested in pure research without product constraints, or if you prefer working on problems where success is measured only by academic metrics, this role probably isn't for you. We need someone who genuinely cares about user impact and is willing to prioritize practical improvements over theoretical elegance when the two conflict.