Bryce Judy’s company began with an occupation he had never imagined wanting. Customs classification is the business of deciding what, precisely, a thing is after it arrives at a border. A shoe is not merely a shoe. Material, construction, use and origin can send it toward a different branch of the Harmonized Tariff Schedule, where a ten-digit code determines duty and points toward other obligations. It is taxonomy with invoices attached.
Judy entered that world through KPMG. Recruited around college, he worked in the firm’s trade and customs practice, classifying products and helping manage the front end of customs clearance. He has described the field as one he had not planned to enter. That accidental introduction gave him an unusually close look at the clerical weight borne by importers, brokers and consultants: government schedules, precedent rulings, changing rates, technical product documents and decisions that need to survive scrutiny.
The work was specialized, but its friction was easy to recognize. Experts spent time searching, comparing, interpreting and documenting. Their conclusions affected the price of goods and the risk of moving them. Repetition sat beside judgment, which is often where useful software ideas wait without making a fuss.
“This was a world that I had never planned on getting into.”Bryce Judy, describing his route into customs
A meeting at a coding competition
At Utah State University, where Judy studied Information Systems, he met computer science student Tanner Helms at a coding competition. The setting matters because Tarifflo would need both sides of the table: someone who understood the trade workflow and someone able to help turn that knowledge into a product. They began working on a system to automate tariff classification, the narrow and troublesome decision Judy knew from KPMG.
The founding question was admirably plain: could they automate what his team had been doing? The answer was not a chatbot with a customs-themed personality. A classification has to connect product facts with the schedule, explanatory notes and prior rulings. An importer needs more than a fluent guess. It needs a result that can be checked, revised and defended.
Judy eventually showed an early version to a former boss. The reaction persuaded him that the idea had moved beyond an interesting demonstration. He quit his job, stepped away from school and worked on Tarifflo full time. It is an old founder ritual - show a practitioner, watch the eyebrows - performed in a field where even the nouns require footnotes.
The route into Tarifflo
The code underneath the code
Judy’s technical work before Tarifflo makes the turn less abrupt than it first appears. Through a project called Pendien, he worked on machine translation for low-resource languages, including English-to-Q’eqchi’, a Mayan language. He appeared twice on the AI Rebels podcast: first to discuss whether language models could support rare and underrepresented languages, and later to explain Tarifflo.
Machine translation and tariff classification are different problems with a shared irritation. Both ask software to map ambiguous language into a precise output without losing context. Both become difficult when examples are scarce or uneven. And both expose the cost of a system that sounds certain while being wrong. Judy’s public work keeps returning to this border between language and rules.
In November 2024, he published “Benchmarking Harmonized Tariff Schedule Classification Models.” The paper proposed a framework for comparing tools on speed, accuracy, rationality and code alignment. It evaluated systems from Tarifflo, Zonos, Avalara and the World Customs Organization’s BACUDA project. Publishing the comparison planted a useful flag: claims about classification should meet a common test.
He added another form of credibility by passing the U.S. Customs Broker License Examination in October 2024 and later receiving his license from Customs and Border Protection. In a young software category, the license is not decorative alphabet soup. It signals that Judy chose to learn the regulated craft his product intends to reshape.
A prize, then a wider product
On March 22, 2025, Tarifflo won the $20,000 grand prize at the Tim Draper Utah Entrepreneur Challenge. The company was still described as a student startup. Its pitch was already larger than a clever code finder: product information went in, and referenced classification documents came out in minutes rather than hours or days.
The timing brought attention. Tariffs had migrated from specialist conversations to daily headlines, and importers were trying to understand how policy changes might alter sourcing, landed cost and compliance. Judy became a translator between those headlines and the machinery below them. On the Ecommerce Finance Podcast, he walked through de minimis rules, duty drawback, customs value, classification and legal tariff engineering. His tone was practical. Be intentional, document decisions, and do not confuse a lower bill with a lawful strategy.
Tarifflo also started moving past the classification wedge. Its current description presents an infrastructure layer for customs: filing software, tariff-recovery audits and live intelligence covering HTS codes, Chapter 99 programs, exclusions and rate changes. The company says it can find entries where importers overpaid because of classification, valuation, origin or tariff stacking. Filing is offered without the per-entry and per-user fees associated with older systems, while Tarifflo makes money when it recovers savings.
That model changes the sales conversation. A legacy tool can charge for access to the process whether or not the process improves. Tarifflo’s recovery pitch begins with a result that can be counted: duty paid, an error found, a claim prepared. Its filing pitch removes a familiar tollbooth and asks companies to work under their own filer codes. Both depend on the same underlying intelligence staying current. Rates change, exclusions expire, special programs stack, and a code that looked complete in isolation can become wrong when another rule enters the calculation.
That expansion follows Judy’s diagnosis of the industry. Getting the HTS code right is one decision inside a much larger entry. He has written that a full entry can contain more than 2,000 fields, including data demanded by participating government agencies, with validators and relationships among them. Classification may be the front door. The house is paperwork all the way down.
A full customs entry can contain more than 2,000 fields. The opportunity is not one clever answer. It is making the answers work together.
The founder as licensed practitioner
There is a recognizable style in Judy’s projects: enter a dense domain, build a test, and keep the evidence visible. In March 2026, away from customs, he entered OpenAI’s Parameter Golf challenge, which asks participants to train a language model in under ten minutes on eight H100 GPUs and fit the artifact inside 16MB. Judy described assigning one agent to scan research papers and another to run training experiments. He reached a strong provisional position, ran out of compute before a required third seed and published the work as a non-record submission.
The episode is revealing because the unfinished label remained attached. In trade compliance, that instinct is useful. A plausible answer without provenance is a liability wearing nice syntax. Tarifflo’s bet is that automation can move quickly while remaining tethered to rulings, schedules and product facts.
By summer 2026, Tarifflo had joined Y Combinator’s S26 batch. YC listed the active company in Salt Lake City with a five-person team. Around the same period, Judy was discussing supplier graphs, sanctions screening, geopolitical risk signals, foreign-trade-zone analysis and a rebuild of customs-entry work. The product surface was spreading, but the founding idea remained intact: complicated trade decisions should be faster to make and easier to inspect.
The supplier work pushes the company one step earlier in the journey. Customs begins at the border, but risk can begin several suppliers away. Judy has described tools that trace deeper supplier tiers, screen names across sanctions programs and monitor country-level signals. The common thread is a graph of relationships rather than a single form. A shipment connects a product to a maker, an origin, a route, an importer, a classification and a shifting set of rules. Tarifflo’s widening scope suggests that Judy sees those connections as the product, not incidental metadata around it.
The useful lesson in the dull aisle
Founders are often told to look for pain. That advice becomes more useful when made specific: look for skilled people repeating expensive work inside a rule-bound system. The problem may not announce itself as technology. It may look like a spreadsheet, a government portal, a bill from a broker or a colleague who knows which obscure document settles the argument.
Judy found his version in the tariff schedule. He then accumulated the pieces needed to build credibly around it: consulting experience, a technical co-founder, a public benchmark, a professional license and repeated conversations with importers. The sequence is less cinematic than a garage revelation and more transferable. First learn where the costly judgment lives. Then decide which part software can carry without pretending the judgment disappeared.
His stated ambition is a one-stop shop for trade compliance and supply-chain management. Whether Tarifflo reaches that breadth will depend on thousands of small, unromantic details. Customs has no shortage of those. For Judy, that appears to be the attraction. The dull aisle is crowded with problems, and almost nobody is posing for photographs there.