Building Turbopuffer: Gergely Orosz (@pragmaticengineer ) × Simon Eskildsen (CEO)

Gergely Orosz worked at Shopify from age 18 to 21, gaining extensive experience across various infrastructure components like caching, storefront development (alongside Justine), and database scaling. He left seeking new learning opportunities after feeling he'd maximized his growth within the company. A significant event during his time was the surge in traffic caused by celebrities like the Kardashians launching products on Shopify, which exposed scalability challenges.

While at Shopify, Gergely initiated a project called 'Napkin Math,' documented on GitHub. This involved creating a table detailing the bandwidth achievable with various hardware components (DRAMM, S3, NVME SSDs, EBS volumes) and their associated costs – including spot instances and committed usage. The goal was to have readily available data for quick calculations and to challenge assumptions in infrastructure decisions.

Gergely frequently encountered situations where product teams made infrastructure choices based on what he considered flawed benchmarks. He felt these benchmarks often failed to account for distributed query complexities, leading to inaccurate performance expectations (e.g., a benchmark showing 10 seconds versus his calculated 10 milliseconds). He developed the 'Napkin Math' project as a tool to quickly perform calculations and challenge those assumptions.

After leaving Shopify, Gergely began exploring the performance of MySQL writes. He initially hypothesized that the number of writes per second should correlate with the frequency of f-sync operations to disk. This led him down a rabbit hole investigating batching and ultimately uncovering that MySQL could achieve significantly higher write rates than initially anticipated due to optimizations like larger f-sync sizes.

Gergely Orosz initially developed a vector database as an internal SAS project, prioritizing reliability and high availability. He released it publicly with the claim of offering 'a million vectors for a dollar,' significantly undercutting existing solutions which cost around $100 per million vectors at the time. This aggressive pricing strategy demonstrated his confidence in the system's efficiency and scalability.

According to Gergely, Cursor’s founders had a dinner table discussion about the inefficiencies of their current vector database setup. They were grappling with the fact that all vectors resided in DRAM and explored the possibility of leveraging S3 for storage and caching, combined with active code bases in memory and other data in object stores. This highlighted an opportunity for a more cost-effective and scalable solution.

After Gergely’s in-person visit to Cursor, Justine, his co-founder, quickly implemented a file-based cache, replacing the existing reverse proxy enginex cache. This optimization significantly improved performance by bringing data directly into RAM for faster query processing. Her ability to rapidly identify and address inefficiencies demonstrated her expertise and contributed to Turbopuffer's adoption.

Gergely and Justine’s work resulted in a remarkable 95% reduction in Cursor’s database bill. This substantial cost saving was achieved by optimizing the infrastructure and leveraging Turbopuffer's capabilities, showcasing its potential for significant economic benefits for businesses relying on vector databases.

Gergely Orosz initially doubted if Turbopuffer could become a billion-dollar company, particularly in the early stages. He described it as feeling like a very niche product and a specialized search engine, which led to a cautious approach rather than immediate aggressive scaling. This initial assessment was based on his understanding of the market and the challenges involved.

The founders prioritized minimizing expenses, specifically focusing on reducing their Google Cloud Platform (GCP) bill. Gergely emphasized a pragmatic approach where the goal was to ensure one number (expenses) remained lower than another (revenue). This demonstrated an early commitment to financial discipline and sustainable growth rather than chasing rapid expansion.

When seeking initial funding, Gergely openly communicated his intentions and limitations to potential investors. He explicitly stated the desired investment amount ($700K) for hiring two engineers and a small buffer, coupled with a clear condition: achieving product-market fit (PMF) within a year or shutting down without significant financial loss. This transparent approach was perceived as unconventional and even concerning by some West Coast VCs.

Gergely outlined three primary reasons companies seek venture capital funding. The first is to fund Research and Development (R&D), which was Turbopuffer's initial goal with the hiring of Buen and Morgan. He then contrasted this with less desirable motivations, such as fueling growth or satisfying founder ego, cautioning against the dilution of employee equity and potential for a status-driven approach.

Simon recounts his early fascination with computers stemming from Microsoft PowerPoint's animation features. He discovered Front Page, a tool intended to simplify web development, which initially seemed promising but ultimately revealed the complexities of cross-browser compatibility issues prevalent in the early 2000s. This experience led him down a path of learning HTML, PHP, and eventually exploring more advanced programming concepts.

Simon humorously attributes his proficiency in English to a four-year period spent playing World of Warcraft. He explains that the game necessitated understanding and communicating in English, significantly improving his language skills. This anecdote highlights an unexpected benefit derived from immersive gaming experiences.

Simon describes his participation in the International Olympiad in Informatics, an experience that exposed him to algorithmic problem-solving. He explains that these problems differed significantly from typical web development tasks, requiring a more mathematical and logical approach – exemplified by challenges like optimizing package distribution across trucks. This competition broadened his programming skillset beyond front-end development.

Simon details how Shopify recruited him while he was still in high school, a surprising turn of events triggered by an article he wrote about the drawbacks of smartphones and his return to a Nokia brick phone. The article gained unexpected attention on Hacker News and was subsequently featured in The New York Times, leading to contact from a Shopify recruiter who was impressed by his insights.