Rokit was founded in Seoul in 2012 by Seokhwan You with what first reads as a science-fiction premise: using a patient's own tissue and cells, combined with AI-driven modeling and clinical-grade 3D bioprinter, to regenerate human tissue and organs at the point of care.

I had met Founder You years before Rokit existed, while he was an executive at another biopharmaceutical company. I was a young aspiring science student at the time, encountered him at a conference, and we fostered a relationship over the years that followed. 

When I later heard that his new venture Rokit was raising, I engaged immediately and wrote one of my very first angel checks — before I truly understood what angel investing entailed. This was still when I was just beginning to be exposed to the world of startups. It was an investment made entirely on my knowledge of the founding team and my personal passion for regenerative medicine, which at the time was widely regarded as too early, too speculative, and close to science fiction.

Rokit's initial product was a clinical-grade bioprinter kit that produced clinical scaffolds from autologous bioinks. As institutions and research groups began using the kits, Rokit accumulated a proprietary dataset of clinical outcomes and eventually built an AI-driven research, optimization, and modeling layer on top of that knowledge base. A major commercial application has been tissue regeneration for diabetic foot ulcers, a condition that often ends in amputation with a high cost burden across the globe. The platform has since expanded into cartilage regeneration for osteoarthritis and a kidney program for chronic kidney disease, and is now in clinical use across more than 20 countries.

Regenerative medicine is a field where the underlying science has been evolving for almost two decades, but where productization has consistently been the bottleneck. I believe that what made Rokit unique was its vertical integration: grounded in differentiated hardware with a software and data layer built on top, which is the structural moat I still look for in defensible deep-tech businesses. For instance, it is the same moat I had later recognized in another investment: Branch Technology, where a hardware breakthrough in large-scale 3D printing was made commercially defensible by the software and design layer built on top. 

After more than seven years since my angel investment, Rokit completed its KOSDAQ IPO in 2025 and reached unicorn status. This was the first IPO outcome of my direct angel investment, and it felt, to me, like one of the most meaningful (and longest) scientific studies over the years. 

Looking back, Rokit taught me a lesson that took years to fully appreciate: not all conviction comes from analysis. Sometimes conviction naturally builds from spending enough time around exceptional people that you develop trust in long before the market recognizes it.

Flex was founded in 2020 by Zaid Rahman, originally as Flexbase: a credit card and billing product purpose-built for construction teams. Construction operators are mostly small and mid-sized businesses (SMBs), and the banking infrastructure available to them at the time was limited, fragmented, and unsuited to the unique payment cycle in which cash has to flow between the project owner, the general contractor, and a long chain of subcontractors.

For me, the original Flexbase vision fit cleanly into a personal thesis: that the financing cycles of legacy industries were structurally underserved, and that vertical software paired with specific purpose-built financial tools was the right model for solving them. It was, in many ways, the first vertical-specific thesis I had developed, and Flexbase was my first angel investment in the construction sector. I clearly remember Zaid, another Columbia alum and a Thiel Fellow, for his enthusiasm, his expertise in construction, and his special way of reducing complex, difficult goals into simple, tactical steps.

Several years on, the team recognized that the underlying pain they had been solving for construction operators was not only a construction problem but one shared by SMBs at large. Cash-flow constraints, payment automation, and access to credit affected the mid-market, and the team broadened the scope of their product accordingly. The launch of the Flex Business Credit Card became the catalyst for evolving a vertical product into a horizontal financial platform; the product now bundles expense management, bill pay, AI-driven accounts payable automation, and high-yield banking into a single cohesive ecosystem.

Flex today is best described as an "AI-native private bank" for middle-market business operators, who employ roughly 40% of American payroll but were forced to run their financial life across ten or more disconnected systems. The company has now surpassed $10 billion in annual payment volume, raised over $480M in combined equity and debt, and grown into one of the fastest-scaling fintech operations in the SMB segment today. 

For me, personally, Flex has been the clearest case in my portfolio of a study that revealed more than any of us had originally anticipated. A vertical thesis had expanded well beyond its first scope, a market had spoke its own will, and a team had the will to listen and the insight to adapt.

Longview Restoration was founded in 2025 by Joey Kim — a dear friend, longtime colleague, my first manager when I entered the VC world in 2019, and a personal mentor I have worked closely with over the years since. The company is special to me for that reason alone, but it is also special on its own right: Longview is building the first AI-powered, full-stack restoration company in what I believe is one of the most overlooked, most fragmented, and most rapidly growing essential-service industries in the country.

Restoration is a genuinely unusual industry, and one that very few people think carefully about. It is roughly a $200B market in the United States, almost entirely run by local operators on legacy processes, and it touches millions of American households at the worst moment of their year — after a fire, a flood, a storm, or a burst pipe in January. A notable feature of the work cycle is that most of the bill is paid for not by the property owner but by the insurance carrier, and the resulting payment cycles are delayed by months and structurally punishing for the operators who deliver the service on the ground.

Restoration was a market thesis I had originally sourced with Takeoff Capital and EquipmentShare, for which the fragmentation of the industry was an opportunity rather than an obstacle. We invested in Albiware (YC S22), an end-to-end restoration management software company that grew very quickly inside exactly that dynamic. 

After several years of working closely with the Albiware team and experiencing the industry up close, it became clear that the real opportunity in restoration was far greater than the software layer alone: that the durable economics lived in owning the value chain end-to-end, with an operational and data moat built on top, in much the same way EquipmentShare had done with the construction equipment industry. This led to the founding of Longview, which uses proprietary AI, operational, and physical workflow automation to drive fundamental advantages. This was perhaps a year before the "AI roll-up" language that leading AI Labs (Anthropic, OpenAI) and larger Private Equity have since helped popularize, but the underlying instinct was parallel.

Joey is one of the most thoughtful, organized, and hardworking colleagues I have had the privilege of working with. And becoming a part of his Longview journey as an angel has been a very meaningful moment. Watching him cross over from the investor side of the table to the founder side has been inspiring. The exciting, early results followed quickly: realized revenue in the eight figures within the first year of operation.

I believe that Longview is, in many ways, the natural evolution of a vertical-specific thesis I have been developing for years. In certain industries, software distribution is not the endgame, but ownership is. The greatest value may accrue to companies that combine technology, operations, and direct participation in the underlying market, rather than to those that simply build tools for it. Sometimes the logical conclusion of understanding an industry deeply enough is not to keep building software for it. It is to step inside, and become part of it.

EarnDLT was founded in late 2019 by Aaron Lohmann, whom I first met at a founder event and remember for the same thoughtfulness, intellectual rigor, and genuine passion I had encountered in my best academic colleagues during my research years. We’ve built a very meaningful relationship over time. I invested as an angel not long after and have been able to watch the product evolve hands-on through the years.

The thesis behind EarnDLT is simple yet profound: energy is the backbone of all modern industry, and the essential “provenance data” describing how that energy is produced, transferred, and legally accounted for has been absent in any tokenized, machine-readable, audit-grade form. EarnDLT is building the data infrastructure layer that fills exactly that gap: digitizing the provenance, transfer, and emissions data of energy across the full value chain so that producers, operators, and buyers can finally transact on a common substrate.

The data infrastructure thesis underlying EarnDLT is personal for me. It is, in essence, the same thesis we helped adopt at EquipmentShare: repositioning the company not as an equipment rental marketplace but as a construction technology and data telematics business. With EarnDLT, we positioned the company as a data infrastructure layer that touches the entire value chain, rather than as a marketplace, a registry, or a software point solution.

The reason this category has not already been built is that energy markets are structurally opaque. Molecules change hands across producers, traders, pipelines, utilities, and end-buyers, and visibility is lost at almost every step of the chain. Exchanges publish price, not provenance. Regulators publish macro aggregates on delayed cycles. Pipeline operators see only their own assets. The work of digitizing this chain at scale has not been accomplished by a single product but by a long-running infrastructure build, led by EarnDLT. 

The compounding asset underneath all of this is the Energy Intelligence Graph: a verified, delivery-level data layer that traces every molecule from production through transport to final consumption. On top of this graph, EarnDLT plans to deliver products that the next decade of energy procurement will require: resilience scoring for data centers, agentic energy procurement and transactions, and the operational intelligence layer that hyperscalers and large consumers will increasingly need as energy becomes the binding constraint on the AI buildout.

Today, the platform is being used by over 40 of the largest energy producers, operators, and consumers across the US and EU, positioned at the center of three converging tailwinds: data center buildouts, US-EU LNG trade, and renewables tokenization.

This is, ultimately, why I find EarnDLT compelling. What began as a data infrastructure thesis is evolving into something much larger — the digitization of energy at scale, made possible only through the quiet, slow construction of the foundational layer beneath. Data infrastructure companies may often look narrow, technical, and slow in their early years, because they are solving foundational problems from the ground up, the right way. But once enough systems are built on top of them, they become extraordinarily difficult to replace. There is a Korean idiom that I recall here: the most valuable infrastructure is the kind nobody notices until it is missing.