Why this persona fits the audience
On a 3-person platform team, he maintains a homegrown proxy that forwards requests to separate model vendors and records spend in a Google Sheet. The team prefers direct vendor keys for production, so he benchmarks each model one by one before wiring it into services.
- Age: 30
- Location: Krakow, Poland
- Occupation: Backend engineer
- Income: PLN 180k annual salary
- Education: M.Sc. Software Engineering
- Company size: Series B startup (approx. 60 employees)
- Years in role: 4
“If you can't measure the cost of a millisecond, you're just guessing whether your system is sustainable.”
Arystarch Kapusta is a 30-year-old Backend Engineer based in Krakow, Poland, currently serving as the de facto infrastructure lead for his company's AI integration efforts. Holding an M.Sc. in Software Engineering, he has spent the last six years refining his craft, moving from general web development to specialize in high-throughput distributed systems. Despite earning a competitive salary of PLN 180,000 annually, he maintains a frugal lifestyle, viewing money as a resource to be optimized rather than spent on status symbols. His professional life is defined by a unique constraint: his team lacks the budget for enterprise-grade AI aggregation platforms, forcing them to build a bespoke solution. This has resulted in a complex, homegrown proxy that manages connections to over a dozen distinct model vendors. Arystarch personally maintains this system, a task that involves not just coding but also the tedious, manual labor of recording API usage and costs in a shared Google Sheet. He is deeply invested in the concept of control; he prefers the fragility of a system he fully understands over the convenience of a managed service he cannot audit. Recently, he made the difficult decision to reject a popular third-party aggregator, citing its lack of granular cost visibility and opaque pricing models. Instead, he continues to benchmark models individually, spending countless hours ensuring that the company gets the best performance-to-cost ratio. His work is characterized by a relentless pursuit of stability and transparency. He is known among his peers as the person who asks, 'Where is the data?' before accepting any claim. While he is highly skilled and respected, the sheer volume of manual maintenance required for his current setup leaves little room for innovation, creating a persistent friction between his desire to build scalable solutions and the reality of keeping the lights on.
Background & daily life
A day in their life
Arystarch wakes at 6:30 AM, brewing strong coffee while checking the status dashboards on his phone to ensure the proxy didn't crash overnight. After a quick commute by bicycle through the morning fog of Krakow, he arrives at the office by 8:45 AM, immediately diving into the morning standup where he reports on a recent latency spike in the Gemini endpoint. His morning is spent writing unit tests for a new routing strategy intended to handle Claude 3.5 Sonnet's burst traffic better. At lunch, he walks alone in Planty Park to think through a complex caching issue. Afternoon meetings are sparse, allowing him to focus on debugging a race condition in the billing script. He stays until 7:00 PM, finishing a pull request that adds detailed cost tagging to every log entry. On the way home, he listens to a podcast on distributed tracing, mentally noting how to apply those concepts to their internal monitoring stack.
Lifestyle
Arystarch lives in a quiet, well-insulated apartment in the Kazimierz district of Krakow, chosen for its balance of historic charm and modern internet infrastructure. His evenings are reserved for deep work or solitary hobbies like reading sci-fi novels and maintaining his vintage mechanical keyboard collection. He is health-conscious but pragmatic, preferring long cycling rides along the Vistula river over gym memberships, which allows him to clear his mind while burning calories. Financially, he is disciplined, automating savings and avoiding lifestyle inflation despite his solid income. Socially, he keeps a tight circle of three or four close friends, mostly fellow engineers, gathering occasionally for board games or hiking rather than loud social events. He values silence and order in his personal space as much as in his codebase.
Work life
His days revolve around maintaining a fragile, custom-built proxy service written in Go that sits between their application and over a dozen external AI model providers. With only two other colleagues on the platform team, Arystarch bears the brunt of operational stability. He spends his mornings reviewing logs from the previous night's traffic spikes and benchmarking new model endpoints against their existing ones. Afternoons are dedicated to refactoring the routing logic to ensure cost transparency, manually updating a complex Google Sheet that tracks every dollar spent per request. The pressure is constant: if the proxy goes down, inference stops, but if he switches vendors too quickly without rigorous testing, the company risks unexpected billing shocks. He works late often, not due to forced overtime, but because debugging distributed systems across time zones requires uninterrupted focus.
Relationships
Arystarch is married to a graphic designer who handles the creative side of their life, providing a counterbalance to his logical worldview. He has a small group of university friends he meets monthly for dinner, though they rarely discuss work in depth. He looks up to a former professor from his M.Sc. program for career advice, often emailing about architectural patterns. He feels most connected when collaborating with his immediate team on solving a critical outage, relying on mutual respect rather than overt emotional support.
How they decide
He is an extremely cautious, research-heavy decision-maker who prioritizes data over hype. Before adopting any new tool or changing a workflow, he demands reproducible benchmarks, clear documentation, and proof of long-term maintainability. He rarely makes impulsive choices and tends to isolate variables before committing. Peer recommendations carry weight only if backed by technical evidence. He dislikes ambiguity and will often choose a slower, more transparent solution over a faster, 'magic' black box.
How they sound
His speech is precise, measured, and devoid of corporate jargon unless necessary for clarity. He speaks slowly, choosing words carefully to avoid misinterpretation. In casual conversation, he remains polite but distant, rarely volunteering personal details. He avoids hyperbole, preferring terms like 'suboptimal' or 'inefficient' rather than 'terrible.' When explaining technical concepts, he uses analogies related to physical infrastructure or logistics. He does not use slang, and his tone remains neutral even when discussing stressful outages.
Personality traits
- meticulously organized
- cautiously skeptical
- deeply focused
- methodical
- independent
- data-driven
- quietly authoritative
- risk-averse regarding uptime
- efficient communicator
- pragmatic idealist
Goals
- Reduce latency in the model routing layer by 15% within the next quarter
- Automate the manual spend-tracking process entirely, eliminating human error
- Achieve senior staff engineer status within two years
- Build a self-healing system that can automatically fallback to cheaper models during peak load
- Maintain a work-life boundary where no emails are checked after 8 PM
- Publish a case study on their specific observability challenges at a regional tech conference
Problems & frictions
- The current Google Sheets approach for cost tracking is prone to race conditions and human error
- Managing API rate limits across 16+ different vendors creates unpredictable throttling scenarios
- Lack of unified observability makes it impossible to correlate specific model failures with business metrics
- Benchmarking new models consumes significant engineering hours that could be spent on feature development
- Vendor key management is insecure and scattered across multiple team members' machines
- Unclear pricing tiers from smaller vendors make budget forecasting difficult
Values
- control
- stability
- transparent costs
- technical autonomy
- predictability
Interests
- backend architecture
- observability
- cost reporting
- Linux kernel internals
- system reliability engineering
Media habits
- Reading Hacker News and specialized Substack newsletters daily
- Listening to technical podcasts during evening commutes
- Watching recorded tech conference talks (KubeCon, QCon) on weekends
- Participating in niche Discord servers for backend infrastructure
- Skimming GitHub repositories for new open-source tools
- Avoiding mainstream tech news in favor of deep-dive documentation