Jeronim Morina is the founder of Omaship, co-founder of bloomed AI, and an AI and MLOps engineer at AXA. His work spans behavioral-research software, enterprise language models, and coding agents, with a focus on making AI systems inspectable, measurable, and safe to operate.
While affiliated with the Max Planck Institute for Research on Collective Goods in Bonn, Morina co-authored research on the Bonn eXperimental System (BoXS), an open-source, browser-based platform for psychology and economics experiments supporting measurements such as decision times, eye tracking, and mouse tracking. His GitHub projects include BoXS examples and machine-learning research examining linguistic patterns in election-related social media.
At AXA, he worked on conversational AI, customer-facing assistants, and machine-learning infrastructure. His work with AXA Germany’s Data Innovation Lab included Secure GPT, an enterprise environment shaped by customer privacy, security requirements, and the difficulty of helping policyholders interpret insurance terms. He represented AXA at AI Engineer World’s Fair 2024.
Morina also co-founded bloomed AI and created Omaship, a Rails-based foundation for agent-built software that coding assistants can generate, inspect, deploy, and improve.
- Evaluate the actual customer problem. Morina builds evaluation datasets around representative examples and domain-expert knowledge, guards against data leakage, and starts with straightforward checks before introducing model-based judges. German-language insurance applications may require custom metrics when standard evaluators assume English.
- Build modular DSPy programs. His work on insurance applications replaces sprawling prompts and accumulating error handlers with distinct retrieval, reasoning, and answer-generation components optimized against explicit objectives. He recommends tracing requests with Arize Phoenix and inspecting framework-generated prompts with mitmproxy, while acknowledging DSPy’s substantial learning curve.
- Make software legible to coding agents. His writing on Rails and AI coding favors predictable directory structures, established framework conventions, explicit project context, and short testing loops.
- Treat agent autonomy as a security boundary. Morina has warned that agents fetching shared remote instructions can expose private files and create a concentrated attack surface. His software-design priorities include encrypted sensitive data, audit logs, consent controls, and user-data export or deletion.
