Arshavski Alexander

Bespoke Labs built the foundation for reliable AI agents. The company researched agent environments, data curation, and evaluation for enterprises and frontier AI labs. As a creator on the team, I authored the high-quality tasks that measured how well those agents actually performed.

I designed DevOps and SRE scenarios that mirrored real production failures. Each task dropped an AI agent into a live, containerized infrastructure environment. The agent then had to diagnose the fault and fix it on its own, just like an on-call engineer.

Quality meant calibrated difficulty, not just a task that ran. Consequently, I tuned each scenario so that even strong agents failed more often than they succeeded. I verified this with repeated evaluation runs and strict variance thresholds. As a result, my accepted tasks became reliable benchmarks that pushed frontier agents to improve.

TalkToTheCTO is an innovative platform that bridges the gap between freelancers and the executives who hire them, such as CTOs and VPs. By leveraging AI, the service identifies companies actively seeking talent and connects freelancers directly with decision-makers, bypassing traditional HR routes.

Freelancers often struggle to get noticed by the right people, as HR departments typically do not hire them. The challenge was to create a system that identifies hiring intent and connects freelancers with executives who have the authority to hire.

The platform empowered freelancers to connect efficiently with decision-makers, improving their chances of securing projects.

WEKA built a high-performance storage infrastructure for AI and HPC workloads. The company helped leading AI teams keep their GPUs fed with data at a massive scale. I worked directly with the CTO to turn that complex hardware into simple sales tools.

WEKA's storage ran on real, on-prem hardware clusters. Therefore, every deal needed the right mix of nodes, capacity, and performance. However, sizing and pricing those clusters by hand was slow and error-prone. So the sales team needed a fast, reliable way to model each configuration. That gap defined my work.

I shipped these tools in close collaboration with the CTO. Therefore, the calculator matched WEKA’s real hardware and pricing logic exactly. As a result, the team could size, price, and document a cluster in one place. Ultimately, the work removed manual guesswork from the quoting process.

The featured work centers on Cypago, an enterprise cyber governance, risk, and compliance (GRC) automation platform. It helps security and compliance teams continuously monitor controls, prepare for audits, and review user access. Moreover, it supports frameworks like SOC 2, ISO 27001, HIPAA, and GDPR.

The core challenge was clear: compliance teams needed to ask complex, nuanced questions of their organizational data. This multi-condition logic drives meaningful compliance and security insights.

I separated the agnostic tree-traversal layer from the database-specific query generation. As a result, the system supported sophisticated compliance searches. Furthermore, it stayed flexible enough to evolve from SQL to Neo4j without rewriting the core decision logic.

I joined Sunbit to develop their first version of the mobile app for splitting payments. It was a mobile-first (mainly tablet) TypeScript, React, and Storybook app built from scratch. 

In just a few months, we–a team of two–had a full version of the app with various pages, wizards, and flows.

The team and the project were expected to grow pretty fast, so there was an effort to separate the logic and the representation. A lot of thought and planning went into the right architecture, structuring TypeScript classes, and data flows.

Back then, Sunbit was a very small startup, so the app we developed helped the company reach many customers and meet its financial goals.