Beyond the low-hanging fruit of AI notetaking, of which I’ve written a lot about in the last few months, I’m starting to hear from and get demonstrations of increasingly sophisticated agentic artificial intelligence tools for other use cases.

Emblematic of these is Charli Capital and its Charli AI, which last week launched the Smart Deal Finder tool, designed to help identify and evaluate investment opportunities across both the public and private markets.

While it is intended for several different investing professionals, from venture capitalists to chief financial officers and chief executives, the company’s founder, Kevin Collins, has a special fondness for RIAs.

“RIAs are probably the most exciting market we work with,” said Collins, who has been working in and building technology for almost 40 years; from wide area communications technology with Motorola in its heyday, to roles as chief technology officer and CEO at several other tech firms, as well as nearly a decade at GE Digital prior to founding Charli Capital and beginning development on Charli AI in 2019.

“When you get into the private equity, hedge funds, etc. they are still very sell-side oriented and have a big investment in using legacy systems, the buy side, on the other hand, is interesting because they have to move quick, yet finding an analyst for them to work with is time consuming and if the analyst leaves— I understand what they will go through,” he said, explaining his own searches for analysts over the years.

Related:Surge Ventures Continues Buildout of Its SurgOne.ai Platform With MGL Acquisition

While Collins said Charli is happy to serve any size advisor, from individual RIAs to large banks with wealth management practices, his team has purposely targeted mid-market firms.

“We do have a good group of mid-tier RIAs and Charli is helping them scale, it also does help them a great deal with compliance tracking,” he said, calling out the Know Your Product compliance requirements in Canada, in particular. Charli AI tracks more than 2 million companies and while it does take in some of the same crowd-sourced data that is publicly available, it can back that data up with citations from public and regulatory filings.

I applaud the firm for its transparent pricing, published on its website, something you do not often see when talking to technology firms serving enterprise clients. This pricing is complex and I’ll leave it to advisors to study with just a ballpark range here: from the free Starter (no credit card required) package for kicking the tires on a few features, to Professional at $150 per month, to Premium at $1,495 per month for midsize RIA firms and, finally, an Enterprise package where prices are negotiated based on features needed or any customizations.

Related:Boosted.ai, Building Agentic AI for Wealth Managers, Raises $15M

“Our AI has to understand the difference between a bank and a large tech company versus an early-stage startup versus … that’s why we’ve spent the last six years developing the AI,” he said.

In addition to delivering investment ideas based on real world language and prompts, Charli AI and its Smart Deal Finder can surface insights and benchmark a potential investment’s performance, as well as track market sentiment, and has been taught to tease out how companies stack up against competitors, all of these results being delivered in real time.

As Collins explains, what took the lion’s share of development work was building inside the technology the ability to find and analyze complex investing data without the end user having to apply complex filters, or building out intricate prompts or going through “keyword gymnastics,” as he puts it.

“There are a lot of competitors in this space that are trying to tap into the private markets, well, both private and public,” he said, naming off-the-record several competitors that range between relying on fully human-powered analysis to varying degrees of automation.

Related:The Minds Behind the Machines: David Plon, Portrait Analytics

“Agentic, truly an agentic AI, depends on how good the AI is at doing sentiment analysis, our AI looks at it in [as close as we can get it] to human terms and surfacing those signals,” said Collins.

“That’s what you cannot get with a ChatGPT or Perplexity, they are not thinking machines,” he said, referring to off-the-shelf commercial generative AI.

Given the focus on AI, I thought to ask him about his own team and the level of reliance they have on automated tools for generating code for their own product.

“We are not using AI for generating pretty much any of our code, we don’t like it, we’d have to review it all, we don’t know where it has come from,” he said.

“I still code to this day,” he said, adding he had been working on code an hour before our call. He said the company has no freelance developers, and currently has 20 employees, predominantly scientists, including many Ph.D.s and engineers.

“One of our co-founders has code on Mars right now, the other, our chief scientist is a quantum physicist who medaled in the physics Olympics, she’s absolutely brilliant and is a mountaineer,” he said, referring to co-founder Joel Emery and Elham Alipour, respectively.

He said part of the trick in developing this type of AI is identifying patterns and teaching the technology how to spot them in a repeatable fashion. He noted that once you have nailed that difficult problem, things become far easier.

“One of the things AI can’t do is normalize data across everything in the world of finance, we go into state, federal, provincial filings, regulatory filings, small business filings, it is a massive amount of raw data,” said Collins.

The AI has to make sense of it all, and as he noted, the actual quality of the data varies greatly and has to be accounted for as well.

“We don’t ever rely on a human for any of that, just the AI bringing it into the system and letting the AI do the analysis—we have integrations with 800 different systems from EDGAR to 500-page S1 filings, [Charli] can easily digest a million-page document for instance,” he said.

And it was important, he said, to reiterate that “not one stitch of that data” gets trained into Charli’s models or goes out to any of the commercial LLMs. This too, required a heavy lift from a development, time, and infrastructure investment standpoint.