Research Engineer, Post-Training

Job description

# Research Engineer, Post-Training **San Francisco, CA · In Person · Full-Time** _Applying to this role will also allow us to consider you for other research opportunities at Vizcom. We believe the best roles are shaped around exceptional people, not just job descriptions._ ## About Vizcom Vizcom is where design teams at companies like Nike, GM, New Balance, and Hasbro bring ideas from sketch to product. Designers use Vizcom to sketch, render, explore color and materials, work in 3D, and prepare concepts for production. The render itself was never the point. The point is the physical thing that comes after it. We call this **pencil to product**. Vizcom is a Series B company with more than $52M raised from investors including Radical Ventures, Index Ventures, and Nat Friedman. Five years of professional designers working this way has created something difficult to reproduce in a traditional research environment: millions of moments where a trained designer, in the middle of real work, decided what should survive into a product that ultimately has to become real. Those decisions create a uniquely interesting research problem. A designer's preference among several candidates can reflect the generator's style, where they are in the design process, what they are trying to make, and the professional judgment they bring to the decision. Existing approaches don't cleanly separate those signals. Understanding that judgment — and learning how to model it — is the challenge this role will help solve. ## The Role As a **Research Engineer, Post-Training**, you'll build models that help us understand and learn from the judgment that carries a design from pencil to product. You'll work closely with the engineer who built our current post-training stack and contribute to a growing body of research documenting the approaches we've tested, what we've learned, and where we've found meaningful signal. This role sits directly between research and product. The models you train will ship to working designers, and what we learn through research will influence what the product captures next. If your primary goal is research that ends with publication, this may not be the right environment. If you're excited by the idea of a reward model influencing what professional designers see in the product, it probably is. We also believe the strongest results won't come from clever objectives alone. They'll come from excellent engineering: correct training code, rigorous evaluations, reliable pipelines, and experiments we can trust. ## What You'll Own - Help define and execute the post-training roadmap for our models, from research plan through production. - Build reward and preference models using years of professional design decisions. - Explore and apply methods including supervised fine-tuning, distillation, preference optimization, and reinforcement learning. - Build rigorous evaluations and research practices that help us determine when a result is real and worth scaling. - Partner closely with Product to understand where today's signals fall short and shape what we capture next. - Evaluate emerging post-training techniques and determine which approaches are worth bringing into our stack. - Build reliable training and experimentation infrastructure that allows research results to translate into production systems. This is a charter, not a week-one checklist. We don't expect one person to tackle everything at once. Part of the role is helping determine what matters most and in what sequence. ## What Your First 90 Days Could Look Like **Days 1–30: Learn and map** Understand our data, post-training stack, existing research, evaluation methods, and the approaches we've already tested. **Days 30–60: Validate** Produce an initial result using historical data that holds up against our evaluation and reproducibility standards. **Days 60–90: Set direction** Help define the roadmap for moving from historical signals toward a closed feedback loop between our models, our product, and the designers using Vizcom. ## What We're Looking For - Strong programming and software engineering skills, particularly for machine learning systems. - Hands-on experience training or post-training generative models, including diffusion or flow models. - Experience with one or more post-training methods such as supervised fine-tuning, preference optimization, reward modeling, distillation, or reinforcement learning. - Experience designing evaluations and experimentation pipelines you can trust. - An interest in product-coupled research, where research questions are informed by real users and models make their way into production. - Comfort working on ambiguous research problems where the right methodology may not yet exist. ## Nice to Have - Experience building high-performance training or inference systems. - Experience optimizing ML workloads or working with large-scale training infrastructure. - Experience working with preference data or human-feedback systems. - An interest in industrial design, physical products, or the people who make them. Above all, we're looking for someone who is more interested in understanding **how professionals decide** than optimizing for what the internet likes. ## What You'll Get - **A unique dataset:** Five years of professional design decisions, with new signals generated every day. - **Research that reaches users:** The professionals whose judgment you're modeling are also the people using Vizcom. Successful research can reach their workflows quickly. - **High ownership:** You'll have meaningful influence over both our post-training research direction and the systems used to evaluate it. - **Close product feedback loops:** Research insights can directly shape what Vizcom builds and what data we capture next. - **Direct access to the founders:** You'll work closely with Vizcom's founders and technical leadership as we build out the research function. ## Benefits at Vizcom - 100% employer-sponsored medical coverag