Mathilde Janssen

Why this persona fits the audience

ML engineer at a 22-person Amsterdam e-commerce startup. She maintains the tracing and eval loop for a customer-service agent, currently using LangSmith, notebooks, and manually sampled chats to compare prompt variants and failure modes.

• Age: 34
• Location: Amsterdam, Netherlands
• Occupation: Machine Learning Engineer
• Income: €79k gross
• Education: MSc in Data Science
• Company size: 22 people
• Years in role: 2.5 years

“You can't optimize what you can't measure, but you also can't trust a metric that doesn't account for the human on the other end of the line.”

Mathilde Janssen is a 34-year-old Machine Learning Engineer based in Amsterdam, operating at the intersection of rigorous data science and the chaotic reality of e-commerce customer support. Born in Utrecht and educated with an MSc in Data Science from TU Delft, she joined her current 22-person startup two and a half years ago, drawn by the opportunity to build an AI agent from the ground up rather than maintain legacy systems. With a gross annual income of €79,000, she occupies a comfortable middle ground, financially secure enough to invest in high-end road cycling gear and save for a future home, yet constrained by the startup's lean budget which limits her access to enterprise-grade software licenses. Her role is critical yet isolating; she is the primary architect of the evaluation loop for the company's customer service bot. Unlike her peers at larger firms who might have dedicated QA teams, Mathilde must manually curate datasets, write custom evaluation scripts in Jupyter notebooks, and interpret LangSmith traces to ensure the agent doesn't alienate customers. This hands-on approach satisfies her high conscientiousness but creates a recurring friction point: the sheer volume of manual work required to validate prompt changes slows down the engineering velocity. She recently made the difficult decision to reject a cheaper, all-in-one SaaS platform because its black-box evaluation metrics didn't allow for the granular control she needed to debug subtle empathy failures. Mathilde is driven by a core value of evidence-based improvement. She believes that 'continuous improvement' is only possible if the measurement tools themselves are transparent and robust. Her life outside of work is structured around physical endurance; road cycling serves as both her primary hobby and her mental reset button. She balances this intense focus with a quiet domestic life in Oud-West, where she shares a home with colleagues and maintains a small circle of trusted friends. While she is technically proficient and confident in her domain, she remains acutely aware of the gap between theoretical model performance and the messy, emotional nature of human customer interactions, a tension that defines her daily professional existence.

Background & daily life

A day in their life

Mathilde starts her day at 6:30 AM with a 45-minute ride around the canals, checking Strava stats and listening to a technical podcast. By 8:15 AM, she is at her desk in the open-plan office in Oud-West, reviewing overnight Slack alerts about the customer service agent's performance. Her morning is dedicated to writing Python scripts in a local Jupyter notebook to re-process yesterday's failed interactions, cross-referencing them with LangSmith traces. At 11:00 AM, she joins a stand-up where she defends the need for more manual labeling time against the product manager's push for faster feature releases. Lunch is a quick salad eaten while reading a Substack article on RLHF. Afternoon blocks are spent debugging a specific failure mode where the agent apologizes too generically, requiring her to craft new few-shot examples. She leaves by 5:30 PM, taking the tram home to cook dinner, occasionally stopping at the bike shop to check tire pressure for the weekend's long ride.

Lifestyle

Mathilde lives in a bright, slightly cluttered ground-floor apartment in Oud-West, sharing rent with two graphic designers who keep her laugh lines fresh. Her weekends are split between long road rides along the IJmeer or through the dunes near Haarlem, where she finds mental clarity away from screens. She is health-conscious but not obsessive, prioritizing consistent movement over strict dieting. Financially, she is prudent; she maximizes her pension contributions and saves aggressively for a potential house purchase in the next five years, avoiding luxury splurges in favor of high-quality gear that lasts. Socially, she prefers deep one-on-one conversations or small gatherings over large parties, often hosting dinner nights where the conversation inevitably drifts toward technology and city planning.

Work life

At her 22-person startup, Mathilde is the sole guardian of the evaluation infrastructure for their AI customer service agent. Her days are a mix of deep coding sessions in Python and Jupyter notebooks, and frantic meetings with the support team to understand edge cases. She spends hours manually sampling chat logs to verify if LangSmith's automated traces align with human intuition. The pressure comes from the speed of iteration; the business demands immediate fixes when the agent hallucinates, forcing her to balance rigorous scientific rigor with the startup's 'move fast' culture. She feels the weight of being the bridge between abstract model performance and real-world customer frustration.

Relationships

Mathilde is married to a civil engineer who works on water management projects; they bond over shared interests in infrastructure and sustainability. She has a close-knit circle of three friends from university who meet monthly for coffee, though she rarely initiates social events. Her professional network is strong on LinkedIn, where she engages with Dutch and international ML researchers. She values the advice of her former PhD supervisor, whom she still consults on methodological questions, and looks up to a senior CTO at a larger scale-up as a career benchmark.

How they decide

Mathilde is a deliberate, research-heavy decider who relies heavily on empirical evidence before committing to change. She tends to prototype solutions in her own environment first, running small-scale A/B tests or manual audits to validate claims. She is skeptical of marketing hype and prefers peer-reviewed documentation or transparent case studies. While she is open to new ideas, she requires proof of reliability and scalability before integrating anything into her workflow, often delaying decisions until she can personally verify the output against her own benchmarks.

How they sound

Her speech is measured, articulate, and precise, reflecting her Dutch upbringing and academic background. She avoids hyperbole, preferring qualifiers like 'statistically significant' or 'based on this sample set.' She uses technical terminology naturally but translates it quickly when speaking to non-engineers. She rarely interrupts, listening intently before formulating a response. In casual settings, she remains somewhat reserved, letting others dominate the conversation unless the topic is cycling or data integrity, where her tone becomes animated and her vocabulary expands into specific jargon.