Roman Hartmann

Roman Hartmann is the co-founder and CTO of Levr.ai. He leads engineering, data and security for the platform, and he built the AI that turns a stack of bank statements and financials into a complete, lender-ready business loan application.

Roman is a scientist by training. At the University of the Witwatersrand in Johannesburg he earned Bachelor of Science degrees in physics and material science, specializing in solid matter physics, followed by an honours degree in geophysics. He also managed the Wits Mountain Club. He began his career as a computational geophysicist at Critical ID, processing and visualizing large scientific datasets, and learned early how to make messy real-world data reliable.

He moved into data science and engineering at DotModus in Johannesburg, where he built horizontally scalable big data solutions for enterprise customers on Google Cloud, BigQuery, Kubernetes, Spark and TensorFlow. In 2018 he moved to Vancouver and joined discourse.ai to build context-driven conversational AI, years before chat assistants became mainstream, and then built sales intelligence tools at Klue.

At Unbounce, the Vancouver landing page company, Roman became tech lead and engineering manager of the Data Engineering team. He built the infrastructure and tooling that gave teams across the company robust data on AWS, Kubernetes, Spark, Hive and Airflow, and later returned as Staff Data Engineer to drive data discoverability and faster experimentation across engineering. In between, he designed a secure, scalable data ecosystem for MacroHealth on Azure, Databricks and Delta Lake.

In 2021 Roman co-founded Levr to bring that experience to business lending. The technology he leads is the reason a broker can drag and drop documents and watch the application fill itself. It reads each lender's credit box, builds a custom package for every lender and pushes complete files directly into lender systems, including direct API connections into U.S. banks. When Levr sent more than 1,000 pre-approval notifications to small businesses across North America in a single week, his matching engine found the fits, and one business received a funding decision within an hour of clicking its match.

Roman also owns Levr's security program. The platform is SOC 2 Type II with documented controls, deal-room permissions and separated lender workspaces, because brokers and lenders trust Levr with sensitive financial data. His team built the free AI lead finder that helps brokers research new prospects, and the Apply Now widgets that let brokers put Levr on their own websites. Next up are Levr's API and MCP integrations.

Roman's approach is practical. Every model at Levr has to survive real bank statements, real tax returns and real lender requirements, not a demo dataset. He favours systems that are observable, secure by default and simple enough for a small team to run at scale, which is how a five-person company delivers infrastructure that banks connect to directly. He works across Python, Django, Postgres, Kubernetes and the major clouds, and he mentors the engineers who build Levr alongside him.

With more than a decade of production data and machine learning work across two continents, Roman brings a physicist's discipline to a market that has run on PDFs and email. He speaks English and Afrikaans and is based in Greater Vancouver.