Yin Huang, a legal technologist and former patent attorney, presented Project Fungzao to the CodeX group on October 22, 2026. It is a patent portfolio analyzer and data pipeline that starts from a company name rather than a patent number, then groups that company’s U.S. filings into families and draws how they connect.

The problem. Free patent tools are fragmented, and paid tools are expensive and reluctant to release bulk data. Patent Center only searches by individual application or patent number and throttles repeat queries. Assignment Center searches by company name but returns reel/frame numbers, not patents. Google Patents has the best interface but omits unpublished applications, which matter for enforceability and priority dates.

What the tool does. Given a target company, it pulls every USPTO filing for that company and sorts the filings into patent families. It then calculates utility expiration dates, including patent term extensions and terminal disclaimers. It also reports inventor and applicant counts and parent/child priority claims, and draws each family as a color-coded priority graph.

Why families matter. Statements made anywhere in a family must be consistent with statements elsewhere in it, so any dispute over one patent can implicate dozens of others. Yin said building these lists by hand was a standard first-year-associate task that took hours.

The demo. Yin showed pre-downloaded data for nine companies. Revolution Medicines had 72 families, the largest with 14 members. Nvidia’s portfolio had 6,990 families, and one 45-member family showed a very complicated priority structure. A Moderna family ran to about 200 items, which Yin used to show how hard it is to size up a lawsuit over a single patent in a large family.

Use cases and roadmap. Yin sees investment analysis, competitive intelligence, M&A diligence, and independent in-house data pipelines. Features already built but not yet on the dashboard include pending, allowed, and abandoned application statistics, Sankey diagrams of ownership history, claim-level event tracking, and early signals of imminent patent grants. Next steps include adding PCT data and using the USPTO bulk data holdings, which go back to 1976.

Q&A takeaways.

• Finance is the most promising audience. Yin said he has found little patent-specific research geared toward equity analysts, and he invited anyone with connections in that industry to reach out via LinkedIn.
• The tool is industry-agnostic. It works for any company, from pharma and quantum computing to aging-reversal firms.
• AI-native patent platforms mostly serve a single attorney on a single invention. Yin sees portfolio-level analysis as a gap and is open to collaborating with them.
• He uses AI coding assistance for troubleshooting and for complex SQL queries, and says he still needs to go back and verify how some of the generated queries work.
• He is offering to help organizations build their own independent patent data pipelines. He sees independence from vendor decisions about data access, timing, and format as the main value.

Watch Project Fungzao CodeX Group Meeting

Transcript

Roland Vogl: I’m thrilled to have Yin Huang here today. He has presented to this group before. He’s a legal technologist who has worked on a lot of interesting projects, and one deep-dive project he presented here some years ago. He says this is a second or third iteration, or a new direction, of that project, which he has named Project Fungzao.

Yin Huang: I can explain. I do apologize that I don’t have a better name. All the other names I came up with were “Pat” something, and the problem is that every company in IP is called Pat-something, and every company in legal technology is called Lex-something. This one at least promises to be unique.

Roland Vogl: With that, I’ll turn it over to you.

Yin Huang: Great. Thanks, Roland, for that wonderful intro. Hello everybody, my name is Yin. First of all, I apologize that my video is a little dark. There’s light coming in directly behind me and my webcam is struggling, so I’ll try to make things as bright as I can on my end. I have a short slideshow to share with you, just to let you know what the project is. You should be able to see the slideshow, right?

Roland Vogl: Yes, perfect.

Yin Huang: Functionally, this is a patent portfolio analyzer and data pipeline. I’m far from the first person to have this general concept, but I do think the way I’m approaching the problem is different from how a number of incumbents and existing products do it.

You might be wondering what that picture is. It’s a dim sum item: chicken feet that have been fried and then steamed. If you’ve been in or near the Bay Area, chances are you’ve run across them. If you haven’t, I strongly recommend them. They are the iconic dim sum item, and very tasty. After I named the project “Fungzao,” I found out that the correct spelling is actually “fungzaau,” but it’s a little too late to change that. I think the shorter version is easier to remember anyway.

A quick recap of my experience. I got a B.S. in chemistry at Stanford, then a law degree from Columbia Law School with an eye toward becoming a patent attorney. That’s more or less what I did from 2014 to 2019, so about seven years. Then I was an in-house attorney at a startup from 2020 to 2022. From 2023 onward I’ve been in legal technology. My day job right now is sales for a company called Lex Pipe, and I believe the founder, Michael Sanders, is on this call. So, hello. If any of you want to see my LinkedIn profile, the username is yhuang12.

The goal of the project

Yin Huang: Conceptually, the goal is very straightforward. But for reasons that are kind of mysterious to me, there isn’t much in the way of publicly accessible tools that do what my project does.

The goal is simple: I want to make a target company the basis of patent analysis. Given any company, like Google or Nvidia, my project finds all patent filings made by that company at the USPTO, the U.S. Patent and Trademark Office. These are U.S.-specific filings. I’m limited to U.S. filings right now mainly as a matter of data availability; the USPTO is far better than probably any other agency in making its holdings available.

Once those filings are found, the project organizes them into patent families. This is an absolutely essential step. It then summarizes key information and visualizes so-called priority relationships, which are the internal citations and structures that have major consequences for patent attorneys and patent enforcement.

On its face, this is fairly standard due diligence work, and it sounds pretty reasonable. So why is my project necessary?

Why existing tools fall short

Yin Huang: If you look at the tools that are generally accessible, meaning the ones that don’t charge hundreds of dollars per month per user, the landscape is very fragmented. Every tool does one small part of what you need and none of the other things. If you try to manually assemble this kind of legal intelligence about patent portfolios, you constantly bounce between three tools and struggle to reconcile the information.

Patent Center is the official standard USPTO lookup tool. It only lets you search by individual patent or application numbers. If you have a company like Google with tens of thousands of application numbers, what are you going to do, put them in one by one? As an aside, Patent Center also has query rate-throttling issues. Generally, after the third or fourth application number, I get an angry error message telling me to wait a couple of minutes before entering more. For bulk lookups, it just can’t be used.

Assignment Center, another USPTO website, lets you search by company name, which sounds good on its face. But it doesn’t give you patent numbers. All you get is reel and frame numbers of assignment agreements. It doesn’t tell you which patents were assigned to whom and when, so you still can’t figure out who owns what. And there are insurmountable issues with manual data curation.

Google Patents has the best UI overall, but crucially it has no support for unpublished applications. Those are absolutely essential if you want to do anything professionally relevant with patent rights. Unpublished applications have huge ramifications for enforceability and priority dates, things that can cost people millions, if not billions, of dollars depending on which way they go.

Paid search tools are really expensive. In my experience, I’ve worked for two law firms, both patent-focused in their chosen practice area and marketing. Neither signed up for the Westlaw patent-specific search bundle, because they didn’t find it useful enough. I agreed with that decision at the time. It was basically a slightly cleaned-up version of Google Patents, and do you really want to pay an extra hundred dollars a month per person for that?

Finally, one of the big bugaboos of this kind of analysis is that all search tools, paid or free, are really cagey about letting you get the data in bulk. I think the paid vendors understand in the back of their minds that if you can get data at scale, you need them less: you could do one search, get a giant bundle of data, and do whatever processing you want without going through them, which affects their subscriptions. With free tools, I’m guessing a lot of them just don’t want to bother building this kind of feature, because why would they?

Demo

Yin Huang: With that said, it’s time for a demonstration. I have a different screen with the app already loaded, so let me switch to that. Okay, this is what my project looks like. It’s a very simple UI, mainly because the point is to illustrate how the function works. Right now I’ve pre-downloaded data from the USPTO APIs, which give you far more detail than the human-friendly websites do. I pre-downloaded data for nine sample companies.

Revolution Medicines is a good place to start because there isn’t too much data to sift through. You can see immediately how this differs from most incumbent products. When you search for a name like Revolution Medicines, they dump out a giant bag of patent numbers. It’s then incumbent on the user to figure out the structures within that giant list and how the patents relate to each other. One of the key things any patent attorney has to do after getting a list of patents is figure out what the patent families are.

I really think patent families are a drastically under-emphasized aspect of patent law practice. They weren’t talked about in my law school patent class, and even in practice they were begrudgingly mentioned as something you had to do. I believe patent families have to be front and center in any kind of analysis.

A patent family is a group of patents that the applicant, at the time of filing, has declared to be related to one another in terms of content. In practice, that means every statement you make anywhere within a patent family has to be consistent with every other statement you make within that same family. For example, if you file application number one and say the word “widget” in your invention means a screw, you cannot say in application number ten that “widget” means a nail. At a minimum, if you try to do that, you’d better have some pretty good reasons why you’re allowed to make that change. In my opinion, there’s often not nearly enough due diligence about that.

These families are sorted by size in decreasing order. A family of 14 members is not particularly unusual for a medium-sized business or a somewhat larger one; this looks like a startup-stage company. Once you group things into families, you understand that it’s not just a giant bag of patents. Within the bag are 72 distinct families, each representing a block of patent rights that could be enforced against particular other parties or sold to somebody else. Crucially, within each block, all the statements and legal assertions about the technology have to be consistent with one another.

As a practical note, when I was a first-year associate, figuring out what was in these patent family lists was a standard first-year assignment that took hours and hours. I remember chasing down partners, paralegals, and colleagues to ask how you wrangle the patent websites to make this information come out. Thankfully, you can now program things to dump all that information automatically.

The members list is fairly self-explanatory. It’s a list of patent applications, some of which have been published and others of which have issued as actual patents; you can see patent numbers for the issued ones. Cosmetically, I apologize that the column names aren’t as nice as I’d like. I’m one guy working on nights and weekends, and these are the default USPTO field names. They can be made pretty, but I had to give this presentation today as opposed to next week.

Based on these application numbers, you can generate links to outside websites that show the full record, such as the USPTO’s Patent Center or Google Patents for PDFs of the publications and patents.

So far this is qualitatively the same as what you could in theory figure out from Google Patents. But as we move to the right, things become a lot more difficult to figure out. One is utility expiration date, which I calculate automatically. If you’re a patent attorney in the U.S., you know expiration date calculations are not easy, because to calculate the expiration date for any given patent you need to know the legal status of every other patent in the same family. Finding that out manually is a huge chore. Assuming the USPTO data is accurate and I haven’t screwed anything up with priority claim analysis in my code, these should be the exact correct dates for patent expirations.

They also take into account things like patent term extensions, so if there’s a plus-131-day extension, that’s reflected here. They also account for terminal disclaimers. Occasionally the patent office requires you to tie one patent’s expiration date to that of some earlier patent, for reasons that are complex. When that’s the case, the expiration date field is blank, because you have to go look at the other patent. Could you fully automate that? Yes, but it gets into a lot more data processing, so for now I’m just leaving the link.

Further to the right are inventor counts. I’ve cleaned that up so you can see who the first inventor is, but also that there are eight total inventors, so seven other people to look at. Then there’s the first applicant name. Usually there’s only one applicant, but once in a blue moon there are dual applicants. That’s usually an indicator that the patent started out at one company and was acquired, or the company was acquired and the acquiring company took over prosecution. These are interesting situations you need to look out for, but they don’t come out in standard searches.

The parent count and child count have to do with priority claims, which are very important, and figuring them out by hand is quite time-consuming. If you’re a patent attorney this is probably self-explanatory. Patent applications can inherit content from prior applications. The earlier application is called the parent and the later one the child. Under various bureaucratic rules, content gets passed from application to application, and the benefit is that the inheriting applications get the filing date of the original application. It’s a kind of bureaucratic time machine, and there are a lot of legal benefits in terms of avoiding prior art and making examination easier.

Here I not only tell you which parent application goes with which child, I also tell you the filing date difference, meaning how many days elapsed between the filing of the first application and the second. One quirk of the USPTO data is that for whatever reason, many of these provisional priority applications show delays of 1,000-plus days. I’m pretty sure something is wrong with the data schema, because you’re not allowed to make this kind of priority claim when the day count goes beyond 365, which is one year. But it’s there.

Now we get to the real meat. This is the thing I spent endless hours building, or trying to build, as a first-year associate: a visualization of which applications were filed first, which followed, and how they connect through priority claims. You can pull one family out and look at it. Here, [tan] means provisional, so this block of patent rights started with four provisional applications, whose content was all merged into one application. Green means already issued. While that was in process, further applications branched out of it. So whatever technical subject matter this family addresses, there are five issued applications. Provisionals never issue, so I’ll ignore those. Yellow means still under examination, so there are three that could still result in more patent rights. And there are two in red, which mean abandoned or expired. I’m guessing abandoned, since these numbers look recent enough that they probably didn’t issue.

This gives you a quick overview. If you’re in-house counsel, or someone looking to acquire a patent portfolio, and you want to know what, say, [RAS inhibitors] involves, you can get a very good idea at a glance. And you can do this for any patent family within a company’s portfolio.

Revolution Medicines has a fairly small portfolio. Nvidia has 6,990 families available, and I think there’s actually a cutoff date on the data here, from 2001 onwards, so this is still a subset of the total. Even so, you can already see a family of 45 members covering [compositions and methods of detecting polymorphic…], whatever that is. You can see a very complicated priority structure. If you got sued on whichever patent this is, there are umpteen other patents you have to watch to make sure the information is consistent.

You can repeat the exercise for any industry you want. There are pharma companies, and if you like quantum computing, IonQ is one of the hot players in that area; you can look at their portfolio. To give you a sense of scale, let me switch to another window where I’ve preloaded another instance of the app.

[Brief audio and bandwidth problems; Yin confirms he can be heard.]

Questions and discussion

Audience member: Is the way into your system always to start with one particular patent and see how the others arrange around it? Say I was an investor who wanted to invest in a particular technology and wanted to know whether there are clusters of patents in a particular area, or an innovation path. Could I enter certain keywords and have it show me different families, so I could get a sense of that?

Yin Huang: The investor-driven analysis of company technology portfolios is probably where this is most useful. For the more conventional use case, where a patent attorney needs to look up a specific patent and figure out what to do with it, there are tools out there that already do that. There are enough incumbents that there isn’t much undeveloped, patent-attorney-specific functionality.

I know I’ve been taking more time than needed, but look at Moderna. They have a family on modified nucleotides, which is presumably the core of their mRNA vaccine technology. If you go down to the chart, it’s about 200 items. Now the problem becomes: if you’re a patent attorney and they’re suing you on whatever this tiny green dot down here is, what exactly do you do? How do you even figure out how big the problem is? You immediately see this is a gigantic problem with all sorts of operational and, bluntly, financial implications for people on both sides of the lawsuit. This is the sort of thing where you need an automated tool. If you’re a law firm and you receive service of process on a patent lawsuit involving this green patent down here, the last thing you want is to realize you have to track down 199 other patents before you can figure out how to respond.

Potential use cases

Yin Huang: I have a few slides left. I’ve talked longer than I wanted, but let me go through them. On potential use cases, one is what you were getting at: investment decision making, meaning the technological situation of a company. This kind of analysis can also support competitive intelligence. If you’re at one company and want to know what patents your competitors have, I can help with that. And operationally, I think an independent data pipeline is something every legal department should look into building.

I understand there are a lot of pre-made products out there. But the nature of every pre-made product is that it makes certain decisions for you: how much data you can access, when you can access it, and in what format it’s given to you. If the chosen method of presenting the data doesn’t fit your needs, you’re stuck. You could petition the vendor for a new feature, but operationally you’re stuck until that feature gets built, which is not a great feeling.

There are also things I can add to the dashboard right now. I’ve already built the data processing for a lot of statistics about pending, allowed, and abandoned applications; I just haven’t connected them to the dashboard. I can also build Sankey diagrams of ownership history, which are flow diagrams of which owner passes a patent to which other owner, and which help you gauge how quickly patents are acquired. There’s also fine-grained event tracking: individual claim allowance events, and potentially claim-by-claim tracking, depending on what the USPTO data allows. I think some of it does. And there’s earlier awareness of patent grants: there are specific events, like issuance notices, that you can detect months before a patent is actually granted. Finally, you can always add more data to this kind of analysis. PCT data is probably the most relevant, and if there’s other data worth adding, why not?

That’s where my project stands. There’s one last slide I almost forgot to share: where do we go from here? Fundamentally, if your organization wants to build an independent data pipeline for patent analysis, we should talk. The value I can provide is the independence of the data pipeline. If you just look at the end result, there are any number of vendors who will give you nice dashboards and visualizations. But the one thing they absolutely won’t let you see is the underlying flow of how the data gets from its raw form to its final presented form. If they’re not doing the kind of collation or visualization you want, you’re stuck.

Roland Vogl: And you would help a company that’s trying to build a data pipeline for their specific needs?

Yin Huang: Yes. For U.S. patent data, a lot is readily buildable and also pretty robust. The data I showed you was pulled on the fly from an official USPTO API that only came online one or two years ago, so this wasn’t really buildable before that. The USPTO has also released bulk data holdings, where you can get clean XML copies of issued patents and patent publications going back to 1976. That dataset is like gold. There’s a huge amount you can glean from it, and 1976 is half a century ago at this point. Basically anything of remote relevance to existing patent rights is probably going to be captured in that dataset, at least as far as U.S. patents go.

Audience member: Could this be helpful for the AI-native platforms? There’s [Harvey and similar tools for legal practice generally, and Patlytics and Solve Intelligence for the patent world]. Is there a way to collaborate with one of those platforms?

Yin Huang: I’m certainly willing to. I think there are aspects of portfolio analysis that those platforms aren’t adequately addressing. Their mentality still seems to be that the modal use case is a specific patent attorney at a specific firm working with a specific inventor who wants to file for patents on an existing invention. They aren’t necessarily concerned with the broader picture: given a company, what is the universe of U.S. patent rights that company owns?

A probably much more compelling use case is in the financial industry. You can envision an equities research firm trying to figure out [what OpenAI owns versus what Anthropic owns], or something similar for any company. I’ve actually spoken with some equity analysts, and as far as I’m aware there’s little to nothing in the way of patent-specific research products geared toward finance professionals. In my mind that’s a much better direction to develop in. So, everyone in this group with connections in that industry, please reach out.

Roland Vogl: We’re pretty much at the end of our time. If people want to reach you and connect, what’s a good way to get in touch?

Yin Huang: LinkedIn is great. If you want my email, I can also share it. Whatever your favorite way is. I’m very easy to find on LinkedIn.

Roland Vogl: I think I put your LinkedIn link in the event announcement. Great. It’s very cool. It seems to me it would be inherently useful just to understand where things are headed. If you see a concentration of patents and patent families, and you have an easy way to get a clear picture of that, that’s valuable information, especially in an M&A or investment context. Hopefully you can advance this in that industry, and we certainly appreciate you coming in and sharing this with us. It’s very cool that this is a project coming out of your evenings-and-weekends work.

Yin Huang: Absolutely. This is certainly a personal ax to grind for me, because, like I said, when I was a first-year associate so much of my time was spent cranking this stuff out manually. That was not a fun time, let’s just put it that way.

Roland Vogl: Can you say a bit about whether you’re using AI to pull the data together?

Yin Huang: I use AI-assisted coding generally, and it’s very good for troubleshooting overall. With these pre-made software packages, there are often obscure bugs, because you said “abracadabra” when the program expected “Alakazam.” There’s nothing particularly interesting about that; it’s pure formalism, and AI is very useful for it.

The back-end data processing relies a lot on SQL, because the data formats provided by the USPTO require a lot of reorganization before you can really use them. I use AI for the more complex SQL queries, and it’s very helpful at writing them. Some of these would have taken me a day or more to figure out on my own, and I just say, “write this,” and it works. I need to go back and verify how it works, because I’m a little shaky on that. But if it works, it works.

Roland Vogl: Absolutely. That’s where we are.

Audience member: Are you focusing on patents for aging and reversal?

Yin Huang: No, I’m not focusing on any particular kind of patent. The great thing about this is that it’s basically industry-agnostic. Given any company, if you have an aging-reversal company you want to look up, I can pull the data for it and show you all the charts for that company.

Audience member: I’d like to know what those patents are.

Yin Huang: Sure, we can do that.

Roland Vogl: Thank you so much again for joining us, and thank you all. We’ll be closing here. Stay tuned for our next session, which might be next week or maybe in two weeks. I’m not quite sure yet, but I’ll send something out. Good to see you all, and thank you again. A round of applause. This is great, we appreciate it.

Yin Huang: Thanks for having me. Thank you so much. Bye.