Staff Applied Scientist (Compliance & Operations)

Turn scientific insight into business impact.

About Bizee

Bizee helps entrepreneurs across the US start and grow their businesses, with services spanning company formation and ongoing compliance. We make complex steps easier to navigate so business owners can focus on building what comes next.

About the role

Help shape how Bizee uses AI to simplify compliance and improve how services are delivered. As a Staff Applied Scientist, you’ll turn open-ended business questions into models and systems for understanding documents, identifying relevant requirements and completing operational work. Your work will help customers get reliable answers and reduce processing time, errors and manual effort.

This is a hands-on role where you will define the scientific approach when the problem, available evidence or measures of success are unclear. You’ll investigate data, develop hypotheses, train and adapt models, design experiments, and work with engineers to bring successful solutions into production. You’ll decide which problems deserve investment and establish whether improvements in model performance translate into better service.

What you will do

• Partner with Operations, Product, Engineering and compliance experts to identify valuable opportunities. Use workflow data, exception patterns and customer feedback to challenge assumptions and define testable hypotheses and measures of success.
• Develop models and AI workflows for document classification, information extraction, routing and operational decisions. Test their effect on completion time, error rates, manual review and cost per completed task.
• Build the scientific foundations for identifying relevant business requirements and answering compliance questions. Work with domain experts to connect outputs to authoritative, current sources and define when the system should abstain or request human review.
• Train, fine-tune, and evaluate models for specific business tasks. Compare traditional machine learning, pretrained models, retrieval-augmented generation, and supervised fine-tuning, choosing approaches based on evidence, data availability, and operating cost.
• Develop reliable training and evaluation datasets. Define labeling guidance with subject matter experts, address gaps and imbalance, and prevent leakage between training and testing. Use synthetic data or active learning where they demonstrably improve quality.
• Create repeatable evaluations of retrieval quality, factual accuracy, task completion and consequential failure cases. Use real Bizee documents and workflows, expert review and validated automated checks. Set release criteria and rerun evaluations after material changes.
• Own the scientific outcome through production and adoption, partnering with engineers to deploy and monitor models, diagnose performance changes, and improve latency and cost. Build feedback loops, track data and model versions, and define when to retrain, revise, or retire a solution.
• Stay close to advances in AI and test the ideas that could improve Bizee’s products. Explain results and trade-offs clearly, document experiments, and share reusable methods that help the wider team make better decisions.

What you need to have

• A track record in applied science or machine learning of defining problems under uncertainty and taking models from experimentation into sustained business use.
• Strong foundations in statistics, machine learning and experimental design, with deep expertise in NLP, information retrieval or document understanding. You can diagnose why a model fails and design experiments to improve it.
• Hands-on experience training and adapting models, including practical work with LLMs. You understand transformer models, embeddings, retrieval, and fine-tuning, and can explain when changing the model is more useful than changing the data or workflow.
• Strong Python and SQL skills, experience with PyTorch or an equivalent framework, and the ability to write tested, maintainable code for reproducible experiments and production collaboration. Use AI tools to accelerate the work while verifying generated code and analysis.
• Experience with noisy documents, limited labels and unevenly distributed failure cases. You can assess data quality, build representative evaluation sets, quantify uncertainty and account for the different costs of mistakes.
• Sound judgment about AI reliability and sensitive data. You can address unsupported outputs, prompt injection, and data exposure, and design traceable results with appropriate review and fallback behavior.
• The autonomy to identify the problem when goals or evidence conflict, prioritize experiments and make practical trade-offs. You communicate clearly in English and establish methods and decisions that help colleagues work independently.
• A master’s or PhD in computer science, machine learning, statistics, or a related discipline, or equivalent experience demonstrating comparable scientific depth and delivery.

Nice to have

• Experience with compliance, legal, financial or document-intensive products, including evaluation with domain experts.
• Experience with workflow automation, process analysis or operational decision systems, including measuring exceptions and manual review.
• Experience with parameter-efficient fine-tuning, model distillation or agentic systems that retrieve information, use tools and complete tasks with appropriate permissions and review.
• Experience operating ML systems in the cloud, or research and open-source contributions that demonstrate original thinking and practical impact.

What we offer

• Competitive compensation 34,000-47,000 PLN/month, stock options, and an annual bonus.
• A flexible benefits allowance to spend on what matters to you.
• An office in the city center and a friendly, non-corporate atmosphere.
• Real ownership and the opportunity to make a visible impact.