Bernardo
Sana

For more than fifteen years, Bernardo has built software at both ends of the spectrum, the sprawling scale of Microsoft and the exposed nerve of one-person startups. As a senior engineer, he has architected front-end frameworks meant to hold up under Fortune 500 weight, while at home he has shipped consumer products from a blank page to something strangers actually use. Lately, that same instinct has turned toward AI, automating the unglamorous work so ideas can move faster into the world.

The habit traces back to 2014, when he spent nights and weekends teaching himself 3D printing and building a product called Fabrify entirely alone, while still working full-time at Microsoft. It failed as a business. But it taught him what carrying an idea from concept to something real actually requires, the design choices, the tedious middle, the parts no one sees. He has thought of himself as a builder first ever since, the title always a footnote.

Bernardo Sana's career began not in front-end frameworks or AI platforms but in the wiring of embedded systems, at Instituto de Pesquisas Eldorado in Brazil. Across three years there, he moved through a string of telecom and embedded projects, the kind of close-to-the-hardware work that leaves little room for abstraction and demands precision at every layer.

The defining assignment came through a joint venture with Motorola: developing the Bluetooth Cordless Telephone Profile for the Razr platform, one of the decade's most recognizable phones. Sana led a team of five engineers on the effort, an early lesson in shipping for a product whose success would hinge on details most users would never notice.

“He led five engineers on the Bluetooth work behind one of the era's most iconic phones.” Motorola Razr project, 2002-2005

Sana joined Microsoft in 2006 as a test engineer on the Exchange Unified Messaging team, writing specifications and building VoIP automation frameworks for the young service. Two years in, he moved into program management, leading feature planning for speech-recognition and voice-driven experiences, work that quietly set a course for much of what followed. He also took the stage at Microsoft TechFest, presenting demos to audiences of more than a hundred customers.

By 2011 he was back to hands-on engineering, now on SharePoint and OneDrive, building internal tools to provision SharePoint Online and serving as a core engineer on a joint project with the Bing team that cut latency between client browsers and SharePoint across Office 365. The path ran from testing to program management to engineering again, but the throughline was quieter: a fascination with speech technology that would resurface, years later, in his work on AI.

“Work that sparked a long-term passion for speech technologies.” Exchange Unified Messaging, 2008-2011

While the day job had him deep in speech recognition for Exchange, Sana spent his evenings on something smaller and entirely his own: a memory game for Windows Phone 8, controlled by voice. He designed the gameplay, the interface, and the speech model himself, with no team and no mandate beyond curiosity.

Memoraniac! went on to win first prize in an internal Microsoft contest and drew genuine enthusiasm from outside reviewers, an early proof that the instincts he'd later bring to founder-track work were already there, quietly, on the side.

In 2014, while employed full-time at Microsoft, Bernardo Sana started building Fabrify on the side. The idea was a marketplace for 3D-printed goods. What followed was a year of nights and weekends spent doing every job a startup usually splits among a team: designing the experience, writing the front end and the backend, and getting it all into production.

There was no team to hand things off to and no round of funding to buy time. Fabrify launched and reached paying customers within a year, entirely on Sana's own effort. It stands as an early marker of the instinct that would define his career: build the whole thing, and build it yourself.

“Built after hours. Shipped like it wasn't.” On building Fabrify alone

In 2015 Sana left Microsoft for Workpop, a Santa Monica startup building a hiring platform. The work was full-stack and front-end-heavy: messaging chat, email services, cron-based workflows, and most of the site's user-facing pages, the kind of broad ownership that a small, growing team demands. Workpop was later acquired by Cornerstone.

Three years later he joined InVisionApp, the design collaboration company eventually acquired by Miro. Working in Golang, ReactJS, NodeJS, and GraphQL, he built end-to-end features, API integrations, and internal tooling for a high-traffic platform, working closely with designers and product managers to ship polished, considered user experiences across the product suite.

“Collaborated closely with designers and product managers to deliver polished, high-quality user experiences.” InVisionApp, 2018-2021

Sana came back to Microsoft in 2021 to lead engineering on the Digital Store platform, building a multi-market solution from scratch for the U.S., Japan, and China. It was his job to take it from concept to launch, coordinating the full stack while mentoring two junior engineers through front-end architecture and shared-component workflows along the way.

The centerpiece was Shop in 3D, a modular web widget for exploring Surface devices, hosted end to end in Azure. React and server-rendered Node handled the front end, .NET C# powered the APIs, and a BabylonJS player rendered the 3D models themselves. Thousands of customers now use it daily, a small, quietly persistent piece of how people shop for Surface.

For nearly two years, Sana worked on a problem that rarely makes headlines but shapes everything downstream of it: how a platform learns what works. He designed the integration between Microsoft's internal Experiences Platform and Adobe Magento 2, giving the Microsoft China Store team the ability to flight features and run A/B tests across multiple experiments at once.

The work sat squarely at the intersection of infrastructure and experimentation, built in PHP and Magento, with ReactJS and .NET C# App Services rounding out the stack. It was unglamorous by design. But the testing framework it enabled became part of the core infrastructure supporting the global launch of Microsoft Surface devices, a reminder that big launches are often carried by small, careful decisions made months earlier.

In his first months in Marketing Engineering, Sana integrated Azure AI Search into Microsoft.com's Customer Stories platform, reshaping how visitors found and engaged with the site's content. Story Clicks rose 810 percent. Search Utilization climbed 527 percent. That same year, he proposed and led a hackathon project to generate AI summaries for Microsoft's search pages, an idea that made it to production and lifted click-through rates by 10 percent.

The work continued with the Marketing Assistant, a tool for turning white papers and eBooks into finished marketing emails and campaign pages. Sana built its AI inference layer using Semantic Kernel, Azure AI Search, and Azure OpenAI, orchestrating a pipeline that could read a source document and write from it. The approach has since been submitted for two separate patents.

“A hackathon idea, shipped to production, became a permanent fixture of Microsoft search.” Marketing Engineering, 2024

Sana's current work began with a rebuild. He restructured the Store Assistant's interface into a shared, theme-aware component platform, replacing duplicated code with one system deployable across Microsoft Store, Xbox, Windows, and Surface. It was a foundation-laying move, one built less for the feature in front of him than for the years of iteration to follow.

The technical gains followed quickly. Feature-flag orchestration and chain-of-thought rendering gave the assistant new flexibility, while a fix to a long-standing networking-cache defect cut bandwidth per request from hundreds of kilobytes to roughly one. A modernized rendering pipeline and mobile optimizations pushed engagement up by more than half.

Increasingly, Sana's tools build the tools. He raised automated test coverage from about 10 percent to 76 percent using AI-driven workflows, then built autonomous agents, powered by Chrome MCP and a suite of Storybooks, that carry tickets from analysis through planning, implementation, and testing. The assistant, once a single feature, is now a scalable, multi-tenant platform readying itself for new markets.

“Architectural improvements and AI-driven automation enabled rapid feature expansion, multi-tenant support, and readiness for new markets and surfaces.” On the Store Assistant's evolution