Investment analysts often jump between documents, dashboards, and research tools, relying on manual methods to form a thesis before investing in any asset class. Even with access to the same data, different analysts reach different conclusions with traditional processes and piecemeal tools that are slow and fragmented.
This is the gap Snow Mountain AI was built to close. Founded in 2023 by Pradeep Ayyagari, Brij Singh, and Nilesh Trivedi, the Bangalore-based company operates as an R&D-focused AI lab developing AI models for risk evaluation in banking, private equity, and soon commercial real estate.
Analysts in these industries can utilize their proprietary reasoning AI Model, called the CORE – Consensus-based Observer (driven) Risk Evaluation, which employs a multi-AI-agent architecture to assess quantitative, qualitative, and scenario-based risk signals across all asset classes.
Snow Mountain’s Chief Product Officer, Pradeep Ayyagari, says, “We started thinking about how to give the analyst one place to upload whatever they have, whether it is a pitch deck, financials, or company documents, and let the system take over. The idea was to make their life easier.”
Experimenting with AI in Unstructured Product Spaces
Pradeep is a seasoned product veteran and no stranger to scale. In his previous stint at Flipkart, he designed the complete onboarding experience for sellers and was instrumental in scaling the platform to over five lakh sellers. At Swiggy, Pradeep was a key resource in helping the platform scale from 200 restaurants to over 30,000.
When ChatGPT launched in 2022 and changed the world, the entrepreneurial bug bit Pradeep, leading to the start of his journey with Snow Mountain AI. The “agent-tech” company initially started off as a broad exploration of the capabilities of LLMs and how to use it to scale an organization’s operations.
Pradeep says, “We wanted to get a good grasp of the technology, what it was capable of, what we could build and what we couldn’t. We started Snow Mountain AI as an experiment. LLMs were already around, and we started exploring whether we could use them to operate tools, like a browser or a product that a company uses, things like that. Early agent tech was broadly what we had an idea about.”
From Banking Reluctance to Private Equity Workflows
Snow Mountain was built alongside Forecast 360, their proprietary product built for US banks to run credit loss models and manage underwriting. But with banks’ reluctance to adopt AI, the team repurposed the risk tech product for private equity firms instead, as they were more open to using AI for research and risk evaluation.
Snow Mountain is structured as a single overarching product with a broader go-to-market focus on banking, along with Forecast 360. It operates independently for now. Pradeep explains that the long-term plan is to bring everything under the Forecast 360 umbrella once banks become more open to adopting AI-driven solutions.
Pradeep says, “Alongside Forecast 360, we built a lot of risk tech as well, but we didn’t get much traction from the banks. Banks in the US were very wary of anything AI. We weren’t even getting meetings when we mentioned it. So we had to completely separate the two products. We ended up building a separate, independent product that was more tailored to private equity since they were more open to trying out AI for their use cases.”
The gap Snow Mountain identified lay in how private equity analysts evaluated deals. While the process was deep and data-heavy, talking to the analysts revealed how there was no standard workflow across firms. Analysts relied on a mix of ad hoc tools like Notion docs, Probe42 and spreadsheets to piece things together.
And even when working with similar data, decision-making varied widely from deal to deal. Snow Mountain AI saw an opportunity to streamline the process and bring structure to how analysis was done.
Pradeep explains “The idea was to give analysts a single workspace. Instead of jumping across piecemeal tools, users can choose to upload everything into Snow Mountain Capital’s platform, where they can input all deal materials and have AI agents assist with analysis, flag risks, and generate insights in one place.”
Snow Mountain AI is built around a multi-agent system, designed to automate and accelerate the kind of research and analysis typically done by private equity analysts. Once a company’s documents are uploaded, different AI agents take over specific parts of the workflow:
- Credit Analyst Agent is responsible for unpacking a company’s financials. It parses through balance sheets, income statements, and other financial reports to build a foundational understanding of how the business is performing.
- M&A Analyst Agent is built to interpret mergers and acquisitions data. It reads through annual reports and regulatory filings especially in the Indian context while raising the right questions an investor would typically ask when evaluating a potential deal.
- Deep Research Agent moves beyond documents and into the broader context. It gathers information on the company’s competitive landscape, tracks its history, and flags any red flags such as lawsuits or regulatory issues that might affect the investment thesis.
Overseeing the process is a lead private equity analyst AI agent that distributes tasks across the others and pulls the findings into a single, structured output. Within 20 minutes, the platform delivers a complete analysis that gets ready for a human analyst to use as the starting point for deeper evaluation.
Pradeep explains, “If an analyst had to do this manually, it would take anywhere from two to four days. That is the opportunity we saw, so we built this multi-agent setup. We are seeing good results. Sometimes it is challenging, sometimes it works flawlessly, but the customer response has been strong.”
Reconciling User Feedback with Actual Behavior
Snow Mountain AI began taking shape in late 2023, with the team already clear about the problem statements they wanted to solve. But as they started building, they quickly ran into the limits of early LLMs. The technology wasn’t mature enough to deliver reliable outputs. As a result, they decided to pause active development on Snow Mountain and focus their efforts on Forecast 360.
The turning point came in early 2025 with the release of models like DeepSeek-V2, Gemini, and Claude 3. These models brought in with the capability in terms of long-context reasoning and task planning, which made it possible to build multi-agent systems that could follow workflows
Pradeep says. “That was a pivotal moment for us. It was the point where reasoning became possible, where you could direct LLMs to perform tasks, think through steps, and apply logic. That kind of reasoning was key to unlocking most of the use cases we see in AI today. Once we saw that shift, we started working on it again, built a product, and launched it with a few clients.”
One of the key insights from early user feedback was the tendency for founders to over-engineer solutions. Pradeep observed that when founders hear a problem from users, they often jump to complex, edge-case scenarios in an effort to differentiate their product, when in reality, most users are looking for much simpler solutions.
The gap between what users say and what they actually do became a recurring pattern during Snow Mountain’s early iterations. Pradeep recalls his journey where a more grounded approach in watching how users currently solved a problem, then addressing the simpler parts first, proved to be far more effective in shaping the product.
Pradeep says, “When you see them act versus when you hear them say, there’s a huge gap in that space. That is something we’ve repeatedly seen happen again and again. We try to solve the most edge cases and the most complex things, assuming that this is something that should happen or this is how my product will differentiate. But in most cases, the simplest of things are what really give the home to the customer also.”
Continuous evaluation and adaptive management of AI-driven agents
Building Snow Mountain in the early stages was pretty straightforward. Much of the complexity that would traditionally require detailed coding was now handled by the model through prompts. This shift made it significantly faster to prototype and build features, but the real challenge emerged after the product was functional.
This was to make the product reliable. Pradeep explains that a large part of the problem was evaluating how well the agents were actually working, especially when outputs were probabilistic and could change with new data, prompts, or models.
“Right now, 80% of our effort goes into evaluating the agents, measuring their impact, benchmarking reliability, and fine-tuning them to perform better. That’s a shift from earlier, when 80% of the time went into building the product and only 20% into testing. Now it's the opposite. And evaluation isn’t a one-time step, it is an ongoing process.”
He says, “The best analogy I can think of is that it’s like hiring a new co-worker. You bring them into the organization, evaluate their performance, assign them tasks, observe their output, and guide them along the way. Based on how they perform, you decide whether to promote them or not. That entire process of grooming a new employee is essentially what you have to do with agents.”
Snow Mountain measures success by tracking how accurately and consistently each agent completes its tasks. This serves as the core indicator of whether the system is delivering value. Pricing is currently based on API and token usage with a small markup for early customers.
Over time, the team plans to shift to an outcome- or effort-based model, where agents could be priced similarly to salaried analysts, depending on the value they generate.
The shift to user-led AI adoption in enterprises
“Earlier, it was only developers who played with new tools. Now, even a marketing person has seen what AI outputs look like. Even a CEO has used an LLM to generate a deck or ask a question. So when we walk in with a product, expectations are already set,” Pradeep says.
"The traditional Figma style land and expansion approach to enterprise is changing. It is no longer about following the classic enterprise sales cycle. That is the biggest shift founders will need to adapt to going forward."
