melange
The company is led by Josh 1 and Glenn 2. Josh 1 met Glenn 2 in the third grade. At their high school graduation, Josh 1 gave a speech as the valedictorian; Glenn 2 gave a speech as the class president. Josh 1 and Glenn 2 matriculated to Brown University together and were roommates throughout their time there. Their most notable college adventure together was running a socialist congressional campaign for their high school biology teacher. They would receive 7.1% of the vote in Rhode Island’s 2nd House District.
While at Brown, in a graduate seminar called “Corporate Aesthetics”, Glenn 2 befriended with Kenneth 3. Their first interaction was in the hallway of the English Department for a different class. Kenneth 3 had in his hands a twenty page essay he had written critiquing “The Game: Penetrating the Secret Society of Pickup Artists”.
After graduation, Josh 1 and Glenn 2 went in different directions. Josh 1 accepted a full time offer to work at Susquehanna International Group (SIG), the world’s largest options trading firm. Glenn 2 decided he would be an entrepreneur. Glenn’s 2 first business entity called “WeFlop, Inc.” WeFlop was a decentralized poker platform where anyone could create and monetizable their own poker room. Beside receiving a ten thousand dollar grant from YCombinator in 2020, WeFlop would receive no outside funding.
While working at SIG, Josh 1 would help out here and there with the business affairs of WeFlop. One weekend, for instance, Josh 1 and Glenn 2 played $4.3 million of online blackjack because DraftKings was running a promotion that gave .6% edge per hand. The money earned went into the WeFlop’s treasury.
Eventually, Glenn 2 would help Josh 1 to quit his job and form a new entity called Melange. Originally, Melange was a prediction markets company. The original idea was to build a product that could allow people to check the probability of important future events the same way people check the weather app for the probability of rain. The name of the company was a reference to Dune where the “melange” is a psychedelic substance that allows its user to see the future. Glenn 2 and Josh 1 raised $9.1 million dollars in a Series Seed for this idea.
This early period of Melange was a difficult time characterized by dysfunction, power struggles, and difficult lessons taught by intense pain. It would take two years and two million dollars burned for Josh 1, Glenn 2, and Kenneth 3 to come to the conclusion that prediction markets were a bad business both ethically and practically. They wound business operations down, consolidating the team to conserve burn. This was a stressful time. Some investors wanted their money back. Other investors convinced us to keep going.
We spent the next four months identifying and validating startup ideas. ChatGPT had been released only a couple months earlier. Over time, we developed a thesis that in the context of the legal industry, the generative aspects of LLMs were mostly useless. Most boilerplate legal writing was already template-based, and what was not template-based was specifically the work a client would want a human lawyer to handle. A better use case is text processing as LLMs demonstrate a remarkable aptitude for reading comprehension at a high velocity.
This thinking led us to the patent system. Patent law, more than any other form of law, relies on huge amounts of highly precise text processing. In a prior art search, for example a patent is invalid if every limitation - every constitutive unit of meaning - claimed by the patent is disclosed by one or more prior art references in combination. This assessment often requires reading hundreds of prior art documents with an incredibly degree of precision.
One of our first hires was Mia 4. Mia realized early on that a patent’s claims could not be parsed accurately, deterministically, and consistently by a language model alone. Thus began a multiyear project to build a syntactic parser that could disambiguate the grammar of patent language. To build this system, we hired three more research grade linguists Nora 5, Deb 6, and Matthew 15. The decision to hire Nora 5 was made within thirty seconds. Deb 6 and Matthew 15 were classmates of Mia’s 4 from the graduate linguistics program at UCLA and came highly recommended.
Glenn 2 met Jon 7 via an introduction made by Melange’s most beloved and powerful investor at a board games night. We found Ian 8 and Qingchen 10 through a platform called Paraform. Paraform is a marketplace where recruiters compete to fill open roles inside companies. The product works remarkably well. Italo 14 we found through a different recruiting agency. Italo initially joined as a contractor but we liked him so much we eventually bought him out of his contract and hired him full time. Blake 10 was referred through Ian 8. They had worked previously at another startup. Megan 12 was referred through Mia 4. They were classmates at MIT. Back when Melange was still a prediction markets company, Kyle 9 was the company intern. Glenn’s 2 nickname for Kyle 9 is “Good Kyle Hunting” because Kyle 9 refuses to leave Austin, Texas. In November 2024, Glenn 2 and Jon 7 found Victor 13while presenting at a patent conference called IP Counsel Cafe. Finding Victor 13 at the time we did was perhaps one of the most fortunate moments in the history of the company as Victor’s 13 experience as a former patent examiner, prosecutor, and litigator would drastically change our approach to the problem and understanding of the customer. Our last hire was Jeff 16, Jeff 16 was a Quality Assurance Specialist (QUAS) in the mechanical devices department at the US Patent Office, meaning it was his job to review the work output of other examiners.