1Mby1M AI Investor Forum: Manu Rekhi, Managing Director at Inventus Capital Partners (Part 3)

Sramana Mitra: So, let’s talk about the type of AI companies that you are investing in. We could do some case studies first of what you have invested in, and if you could share with us in what stage and state you encountered these companies and what is it about them that made you write the checks?

Manu Rekhi: Sure. I can maybe give you a couple examples and go into it. There’s this company called SciSpace. We typically invest in pre-seed or seed. It’s usually a couple of founders. Maybe they have a working product, some early product-market fit or just a PowerPoint slide depending on where we encounter them. We do mentor founders for sometimes for weeks or months just to sort of understand where you know them more than the product. Getting hastily into bed with somebody else is not our style typically. We do like to marinate and understand.

So in this case, the founder was building a very AI native application to help researchers, whether you’re in pharmaceuticals or in a laboratory or in an engineering company. Typically, you would have to go through thousands and thousands of articles figuring out what body of work has already been done, and then trying to figure out how do you build the next piece, right? This is for everything from protein molecules to the next AI models.

There’re about two to three million published articles every year. You have to go through the artles in different languages, sometimes in German or Italian or, French, and so what would take weeks and weeks of work sometimes months, is now getting done in minutes.

Because they’re very finely tuned to research and pieces, their quality of work is 10, 20, 100 x better than any large language models because large language models are very horizontal. This is very specific down to that body of work that’s been researched, so that company ROI is very meaningful. They’ve gone from zero to $8 million in ARR in a very short period of time.

Sramana Mitra: When you look at a company like this that is essentially a vertical AI company, is it built on top of a particular LLM that’s off the market?

Manu Rekhi: They initially built their own large language models, and then they have substituted that for models as they’ve come up to par. So rather than spending your R&D dollars on building those models, they’re focused on more these small language models and tuning on top of that stuff so that they have differentiation.

Sramana Mitra: What is the degree of defensibility? How do you gauge that degree of defensibility when something is built on top of an off the shelf LLM but is focused on a vertical application?

Manu Rekhi: You’re using the large language models off the shelf that are more for once the results are done to be able to publish the content. The actual reasoning is actually being built on the data sets they have and the analysis and all those pieces that are built on top.

So that’s one part. Your proprietary access to data and pieces that the large language models don’t have, so that’s your data mode, right? The other part is about the workflows that the researcher needs. So it’s not just the fact that I need to get that research done. I need to figure out the workflows collaboration with other researchers.

For example, if I have questions on that article that I just read, can I get in touch with the author of that article? A social networking site is also being built underneath it. I need to be able to collaborate maybe even higher on other pieces. So there’s a whole bunch of other defensible pieces that come through. So the actual reasoning and stuff is just one part of a much bigger workflow.

Sramana Mitra: Okay. Let’s do another one.

Manu Rekhi: Another company is LivSyt, a very boring business. They’re a native AI for project management for large construction companies. Imagine you’re building a tunnel or a high speed railway train from Ahmedabad to Bombay, for example. That’s a good example. You’ll have a few hundred project managers that are making sure that every part of the project is actually getting done, right? If one piece of steel is delayed, that slows down everything else. There’re a lot of cost delays and lots of fines involved.

So this company basically replaces all those human project management labor in between. Instead of a 200 person team, you can do the same thing with 10 or 15 people, and the software does the rest. Their last contract was $5 million plus in India. So, the customers are seeing actual real value and they built AI native upfront. It’s not tackling the traditional SaaS software workflows. This reimagines the entire workflow very differently and actually delivers value. They’re getting really good hard, large contracts from places in the world that would never pay that much money.

This segment is part 3 in the series : 1Mby1M AI Investor Forum: Manu Rekhi, Managing Director at Inventus Capital Partners
1 2 3 4