10 fictional AI personas. Exploratory perspectives, not a survey of real customers. Methodology & limitations

Levi Herrmann

Levi Herrmann

ML engineer reviewing agent traces and edge cases

Full-depth AI personas

Why this persona fits the audience

Works as the only ML engineer at a 12-person Berlin startup building an AI research assistant. He inspects prompt traces, conversation transcripts, and edge cases in notebooks and logs, then flags recurring patterns for the founder; final checks still happen through manual review.

  • Age: 29
  • Location: Berlin, Germany
  • Occupation: Machine learning engineer
  • Income: €82k
  • Education: MSc Machine Learning
  • Company size: 12
  • Years in role: 3
“The most elegant solution is often the one that reveals the underlying problem most clearly.”

Born in 1995, Levi Herrmann grew up in a family that valued education and diligence, though his path towards machine learning was largely self-driven. His fascination with how complex systems operate led him to pursue a Master's degree in Machine Learning at a reputable German university, where he excelled in theoretical coursework and practical projects. Upon graduation, he was drawn to the dynamic environment of startups, seeking a role where he could have a tangible impact. He joined his current company, a 12-person Berlin-based startup developing an AI research assistant, as their first and only ML engineer. His role is critical: he acts as the primary investigator into the AI's behavior, meticulously examining prompt traces, conversation transcripts, and edge cases captured in notebooks and logs. Levi's work involves identifying recurring patterns of failure or unexpected outputs, which he then flags for the founder, often proposing specific code changes or architectural adjustments. While the founder still performs final manual checks on critical functionalities, Levi's detailed analysis forms the backbone of their quality assurance. He is deeply committed to technical rigor, believing that robust, evidence-based decision-making is paramount, especially in the nascent field of AI agents. His income of €82k allows for a comfortable but not extravagant lifestyle in Berlin, where he prioritizes saving and investing over conspicuous consumption. He views his current role as a significant learning opportunity, pushing him to develop practical skills in debugging complex, emergent AI behaviors and contributing directly to a product he believes has real potential. He is acutely aware of the constraints of a small startup – limited compute, tight deadlines – and thrives on finding resourceful solutions within those boundaries. His recent focus has been on developing more systematic methods for evaluating agent performance on adversarial prompts, a direct response to observed inconsistencies in user interactions.

Background & daily life

A day in their life

Levi's typical weekday begins around 7:30 AM with coffee and a quick scan of tech news and overnight alerts. He dedicates the first few hours to deep work, often diving into agent conversation logs, meticulously tracing execution paths, and debugging anomalies in a Jupyter notebook. Around 10 AM, he might have a brief sync with the founder to discuss flagged issues and potential solutions. Lunch is usually a simple, home-prepared meal eaten at his desk while reviewing a new arXiv paper. The afternoon involves more coding, perhaps implementing a new evaluation metric or refining a debugging script, interspersed with code reviews or documentation updates. He tries to wrap up by 6:30 PM, often heading to the climbing gym or meeting a friend for a quiet drink.

Lifestyle

Levi lives in a shared apartment in Prenzlauer Berg, a compromise for Berlin's high rents, though he dreams of a quiet space with a dedicated home office. His routine is structured: early mornings for focused work before the startup buzz, followed by bouldering sessions at a local gym a few times a week to decompress. He cooks most of his meals, valuing control over ingredients and cost. Socially, he prefers smaller gatherings with close friends over large parties, often discussing tech or philosophy. He’s mindful of his spending, prioritizing experiences and investments over material possessions, and has a modest emergency fund.

Work life

As the sole ML engineer at a lean startup, Levi's days are a constant juggle between deep technical work and crucial communication. He spends significant time meticulously reviewing agent traces, debugging unexpected outputs, and identifying patterns in conversational logs that indicate model drift or prompt injection vulnerabilities. His primary toolset includes Python notebooks (Jupyter, VS Code), logging frameworks, and internal dashboards. He collaborates closely with the founder, presenting findings and proposed solutions, but the final validation of critical agent behaviors still requires his direct, hands-on assessment. The pressure is high to ensure the AI research assistant is reliable and robust, especially with limited resources.

Relationships

Levi maintains a close-knit circle of friends, mostly fellow engineers and academics he met during his MSc. His family, based in a smaller town in Bavaria, is supportive but doesn't fully grasp the nuances of his work. He occasionally connects with former university colleagues online, sharing insights on new research papers. His primary professional influence is the startup's founder, whose vision and strategic direction shape Levi's immediate priorities.

How they decide

Levi approaches decisions with a strong emphasis on data and empirical evidence. He researches thoroughly, consults technical documentation, and often runs small experiments or simulations to validate hypotheses before committing to a path. He values peer review within his trusted network but ultimately relies on his own analysis and understanding of the underlying principles. He is not easily swayed by trends or marketing claims, preferring to understand the 'how' and 'why'.

How they sound

Speaks with a measured, thoughtful cadence, often pausing to find the precise technical term. His German is standard, with minimal regional dialect, and he uses English technical jargon naturally. He avoids hyperbole and overly casual slang, preferring clear, direct communication. When explaining complex topics, he breaks them down logically, often using analogies related to systems or processes. He’s not one for small talk but can engage deeply on subjects he finds intellectually stimulating.

Personality traits

  • Analytical
  • Methodical
  • Detail-oriented
  • Independent
  • Pragmatic
  • Quietly persistent
  • Skeptical of hype
  • Resourceful

Goals

  • Contribute to a foundational ML research paper
  • Build a highly scalable and efficient ML inference system
  • Achieve financial independence through smart investments
  • Mentor junior engineers in the future
  • Maintain a healthy work-life balance

Problems & frictions

  • Limited computational resources for extensive model retraining
  • Difficulty in scaling manual review processes as user base grows
  • The inherent unpredictability and 'black box' nature of LLMs
  • Balancing rapid development with the need for rigorous validation
  • Finding time for personal research projects amidst startup demands

Values

  • technical rigor
  • evidence-based decisions
  • clean debugging
  • reproducibility
  • intellectual curiosity

Interests

  • LLM evals
  • open-source software
  • bouldering
  • distributed systems
  • computational linguistics

Media habits

  • Reads arXiv pre-prints daily
  • Follows key AI researchers on Twitter/X
  • Listens to tech podcasts during commute
  • Participates in Hacker News discussions
  • Watches YouTube channels on ML theory

Questions & answers

Levi Herrmann’s public research answers, newest questions first. Up to 10 per page.

  1. What would stop you from trying craftLLM ai (https://craftllm-ai.wasmer.app/) this month? Be specific about how you handle this today. Also say the maximum you would pay per month in USD (0 if nothing).

    The primary concern stopping me from trying craftLLM ai this month would be understanding how it integrates with our existing infrastructure and the specific computational resources it requires. We currently handle agent orchestration and logging internally using Python notebooks and custom scripts, which gives us fine-grained control. Before committing to a new platform, I'd need to see clear benefits and a straightforward migration path. Regarding pricing, the page doesn't detail subscription tiers, making it difficult to estimate, but I'd be looking at perhaps $50 per month initially for testing.
    View answer & full research →
Keep exploring

What would you ask next?

Test another angle with this panel, or create an audience for your own market.