The Problem That Costs Small Businesses Billions
For millions of small businesses across the United States, bookkeeping remains one of the most persistent operational burdens. The work is essential: without accurate financial records, businesses cannot file taxes, secure loans, or make informed decisions. That said, it is often a repetitive, time-consuming, and expensive process. Traditional bookkeeping services charge rates that strain small business budgets, while the accounting profession faces a well-documented talent shortage that shows no signs of reversing. The gap between the demand for bookkeeping services and the supply of qualified professionals has widened steadily, leaving small business owners with the choice of paying premium rates, managing their books themselves, or simply falling behind.
Synthetic Intelligences Inc., founded in November 2025 by Bench Accounting founder Ian Crosby and backed by Khosla Ventures, was built to close that gap permanently. Rather than augmenting human bookkeepers with software tools, the company is building an autonomous AI system that replaces the bookkeeper entirely: processing financial data, managing workflows, categorizing transactions, and communicating with customers without human intervention. Ian Crosby’s track record gives the endeavor unique credibility; he co-founded Bench Accounting and grew it into the largest small business bookkeeping service in North America, raising over $100 million in funding and employing more than six hundred people, and then co-founded Teal, an embedded accounting platform that was acquired by Mercury. Synthetic Intelligences marks his third venture in small business accounting, this time with the thesis that artificial intelligence can replace the human bookkeeper entirely. Khosla Ventures, one of Silicon Valley’s most prominent venture capital firms, has backed the company based on this thesis and the team’s ability to execute it.
At the center of this effort and engineering the core systems that make its autonomy possible, is Mateusz Kelner, a twenty-four-year-old product engineer whose trajectory from national programming Olympiad finalist in Poland to co-founding multiple technology companies has prepared him for precisely this kind of challenge.
A Builder Since Childhood
Kelner’s engineering instinct emerged early and developed rapidly. He built his first website at age eleven. By high school, he was competing in Poland’s national programming Olympiads, ultimately reaching the national finals, placing him among the top student programmers in the country. His academic achievements earned him multiple scholarships from several institutions: from his high school, from the local government, and most notably from Credit Suisse, whose Quant Scholarship programme at Wrocław University of Science and Technology carries an acceptance rate of approximately one percent. The scholarship is typically awarded to students in their third or fourth year of study; Kelner received it during his first year, an indication of quantitative and technical ability that far exceeded that of his peers.
Kelner enrolled in a Bachelor of Science in Computer Science at Wrocław University of Science and Technology but made a decision that would define the next phase of his career: he dropped out after his first year to pursue software engineering full-time. He had already secured his first professional engineering role at age eighteen, working as a full-stack engineer at BigShortBets, a decentralized finance startup, where he built a social media platform from scratch using React, Django, and PostgreSQL and scaled it to over ten thousand users. He also implemented the frontend of a decentralized peer-to-peer futures trading market, an early exposure to building complex financial infrastructure that would prove directly relevant to his later work.
Serial Entrepreneur, Repeated Success
What distinguishes Kelner from most engineers his age is not only technical skill but a demonstrated, repeated ability to build businesses from zero to profitability. Before joining Synthetic Intelligences, he had co-founded or led four separate ventures, each of which reached meaningful commercial or cultural milestones.
At PMX Labs Inc., the U.S. incorporated company he co-founded in February 2024, Kelner personally designed and built the entire product a digital booster pack marketplace for Magic: The Gathering collectible cards, using Next.JS, tRPC, and Supabase. He created two proprietary algorithms that formed the company’s core intellectual property: a pack odds algorithm that dynamically sets probability distributions for card pulls by factoring in real-time market prices and expected value calculations, and an order-splitting algorithm that optimally distributes customer orders across multiple independent marketplace sellers, minimizing shipping complexity while maintaining price competitiveness. These algorithms powered a business Kelner bootstrapped, with monthly revenue climbing into the hundreds of thousands and monthly profit reaching the tens of thousands by mid-2025. He also oversaw a social media content strategy that he helped grow from zero to four million views per month. He successfully exited the company to his co-founder in late 2025. “What stood out about Mateusz wasn’t just that he could build software; it’s that he’d built and scaled an entire business to three million dollars in revenue on his own,” said Crosby. “That’s the kind of person you want at an early stage startup. He understands the product and the business – not just the codebase.”
In parallel, Kelner co-founded MagicTown, a Polish collectibles trading company operating worldwide with a focus on Europe and the United States. He grew the business from zero to $250,000 in annual revenue before selling his majority shares in November 2024. The company continues to operate and grow under new ownership, demonstrating the durability of the business model he built.
Perhaps the most culturally visible of his ventures was GRP, an esports organization Kelner founded and led as CEO. Under his leadership, GRP became the most popular League of Legends organization in Poland, amassing thousands of followers across Twitter, TikTok, and YouTube. Kelner raised approximately $30,000 in angel funding for the organization and attracted national media attention, including a television broadcast interview on one of Poland’s top three TV broadcasters’ gaming-focused channels and coverage in numerous esports and gaming press outlets. The organization’s rapid rise culminated in receiving the award for Best League of Legends Organization in Poland in 2023 a nationally recognized distinction within one of Europe’s most competitive esports markets.
A Track Record of Building Intelligent Systems
Kelner’s work at Synthetic Intelligences is informed by a consistent pattern of building systems that combine machine learning with practical product engineering. At Purplemana, where he served as frontend lead for an application with approximately one thousand monthly active users, he engineered a card recognition feature that achieved a seven-times speed improvement over the existing solution by deploying a pretrained YOLO (You Only Look Once) object detection model combined with perceptual hashing directly on the client side. The approach was technically novel: running ML inference in the browser eliminated server round-trip latency entirely, enabling near-instantaneous card identification. The innovation demonstrated Kelner’s ability to identify non-obvious technical approaches, in this case moving computation from server to client and implement them in production contexts where speed and accuracy directly affect user experience. He also created the mobile version of the application, first using Expo and then migrating to Capacitor for improved performance.
Kelner has also contributed to significant open-source infrastructure in the decentralized finance ecosystem. He contributed to the dYdX Explorer, an open-source tool that enables users to view, verify, and browse data within the dYdX Layer 2 network. dYdX is one of the largest decentralized derivatives exchanges, historically processing over $1 billion in daily trading volume. The Explorer enables network participants to independently audit on-chain activity, verify balances, and perform forced actions contributing to the transparency and security of a major DeFi protocol.
The structural similarity between Kelner’s prior work and his current role at Synthetic is notable. Both his proprietary pricing algorithms at PMX Labs and the transaction categorization engine at Synthetic require ingesting large volumes of frequently changing numerical data, applying classification and valuation logic, and delivering results with the speed and accuracy that commercial applications demand. The transition from pricing thousands of individual collectible cards in real time to categorizing financial transactions autonomously represents a natural progression of the same core engineering capabilities.
Engineering Autonomy From the Ground Up
Kelner joined Synthetic Intelligences as the company’s second engineering hire, which at a startup at this stage means something specific: there is no existing codebase to extend, no legacy architecture to maintain, no team of senior engineers to delegate to. Every system must be conceived, designed, and built. Kelner is responsible for the data ingestion pipelines that connect to banks and financial institutions, the extraction engines that pull structured data from financial statements, and the AI bookkeeper engine itself the system that categorizes transactions correctly and helps users recategorize the ones it misses on first pass.
The technical challenge of building an autonomous bookkeeper is substantially different from building a bookkeeping tool. A tool presents data and waits for human decisions. An autonomous system must make those decisions itself, with sufficient accuracy that customers trust it with their financial records. This requires not only robust machine learning for transaction categorization but also elegant handling of uncertainty: when the system is confident, it acts; when it is not, it must surface the right information to the user in the right way to enable rapid correction. Kelner’s work on the user-guided recategorization system addresses precisely this boundary between automation and human oversight, a design challenge that sits at the heart of what makes autonomous AI systems viable in high-stakes domains like financial data.
As sole owner of several critical modules, Kelner’s departure would directly impair the company’s ability to deliver its core product to market. The systems he is building – the data ingestion pipelines, the transaction categorization engine, the recategorization workflows – are not features being added to an existing product. They are the product.
Why This Matters at National Scale
The small business bookkeeping market in the United States is not a niche. There are over thirty million small businesses in America, and the vast majority manage their books through some combination of manual effort, expensive professional services, and inadequate software tools. The accounting profession’s talent shortage means that the traditional solution of hiring a human bookkeeper is becoming less accessible and more costly every year. An autonomous AI bookkeeper that delivers professional-quality financial management at a fraction of the cost represents a category-creating opportunity, and the engineering talent required to build such a system is exceptionally scarce.
The kind of engineer capable of building these systems from scratch is not someone who merely writes clean code. It is someone who understands how to architect entire products, who has built and scaled commercial systems under real-world constraints, and who can operate with the independence and judgment that an early-stage startup demands. Kelner’s background bootstrapping a company to $3 million in revenue, designing proprietary algorithms that powered profitable marketplaces, building and shipping ML-driven features, contributing to open-source DeFi infrastructure, and founding ventures that attracted national media attention and industry awards represents exactly the intersection of technical depth and entrepreneurial execution that this work requires.
Building the Financial Infrastructure of the Future
For Kelner, the work at Synthetic Intelligences is the most consequential of his career so far. The stakes are commensurate with the opportunity. If Synthetic Intelligences succeeds, it will have demonstrated that one of the oldest and most established professional services manual bookkeeping can be fully replaced by artificial intelligence.
At the foundation of that achievement will be the engineering work of a twenty-four-year-old who built his first website at eleven, reached the national finals of Poland’s programming olympiads, earned a scholarship that accepts one percent of applicants, co-founded four companies before turning twenty-five, won a national esports award, appeared on national television, contributed to billion-dollar DeFi infrastructure, designed proprietary algorithms that powered a multi-million-dollar business he built without outside funding, and is now engineering the autonomous systems that could transform how millions of American businesses manage their most critical financial data.