How Hero.io is Producing Advanced AI-Driven Financial Insights

A new ecosystem is using AI to simplify Web3 and DeFi decision-making

AI adoption has surged from 45% in the finance sector in 2022 to a projected 85% by 2025, with 60% of businesses already leveraging AI across multiple functions. This rapid integration of AI extends beyond traditional financial institutions, making significant inroads into cutting-edge sectors like Web3 and DeFi, where the complexity of data and the demand for real-time insights are critical.

Hero.io has established itself as a novel ecosystem of AI-driven Web3 and DeFi tools. Among these innovations are Hero Browser, Hero Wallet, and Hero AI Search, developed by Apta—a company founded by top-tier Cambridge PhD researchers. Hero AI Search transforms intricate crypto data into clear, actionable insights, enhancing users’ Web3 experiences by making them easier, more secure, and more effective.

Vatsal Raina, Apta’s CTO and Cambridge PhD researcher in information engineering with research in natural language processing and generative AI and a former innovator at Meta, provided insights into the unique capabilities and future potential of Hero AI Search.

From a theoretical perspective, the mathematical foundations of modern AI were developed several decades ago. However, Raina explains that the recent acceleration in the adoption of AI has been possible with increasingly efficient hardware to run AI systems and the scale of data that is now available to train them. In particular, he notes that the impressive abilities of the recent emerging large language models have enabled AI to gain attention outside academic circles, and like many other sectors, finance is actively integrating various AI technologies to benefit from their powerful applications.

By having an excellent grasp of the underlying theory for natural language processing coupled with practical experience in developing various AI models, Apta’s founding team has been able to create world-class solutions to challenging problems that Raina explains would be impossible with off-the-shelf models.

When it comes to Hero AI, Raina notes, “Apta builds the technology, and Hero offers the route to market. Apta excels at building domain-specific co-pilots capable of performing complex analyses and delivering actionable insights.” Hero enables Apta to focus on developing the technology for the Web3 and DeFi verticals. This partnership guides Apta’s development of tailored and bespoke AI agents optimized for Hero’s target market.

Hero.io is also leading the application of AI to cryptocurrency. Even though there are several crypto analytics tools on the market, Raina claims that Hero AI Search acts as a natural language interface to a complex system capable of comprehensively deriving insights from an incredible wealth of data. This allows Hero AI search users to query across a large breadth of capabilities. For example, which meme coins have gone up the most in the last 11 days? What does the social media sentiment think about Pluton? How correct has influencer X been in the past with their predictions?   

By transforming complex crypto data into actionable insights for users, Hero.io’s AI tools and platform have access to a large range of crypto data sources. For example, Raina explains, “These sources of data include crypto knowledge from vetted textbooks, signals from the blockchain, historical market performance, coin health algorithms, social media sentiment, influencer videos/podcasts, and live crypto news outlets … By developing specialized AI agents interacting for each data source, Hero’s platform is well informed about various sources of information about the performance of any given coin.” This intelligent system can guide users into whether they should trust a particular influencer’s claims or how likely a new coin is to be a rug pull based on the performance of similar coins historically.

Apta essentially only uses an LLM for human-like communication with the user. Beneath the hood, there is a complex interaction between a multitude of crypto-specialized agents powered by crypto-specific data.

Raina also states that traditional LLMs and Hero AI Search are incomparable. “LLMs are just language models and are not designed to perform highly logical tasks. From a technical perspective, LLMs generate language (one word at a time) autoregressively. The choice of each word to generate by an LLM is achieved by sampling over the distribution of the vocabulary of all words. The distribution of the vocabulary is a direct effect of the type of language the LLM has seen at the time it was trained. Therefore, the LLM only mimics the sequence of words seen on the internet.”

In contrast, Hero AI Search relies only on an LLM to interface with the human user. Behind the LLM sits a complex, crypto-specific agentic framework (“experts”), which feeds the LLM with insightful information, making the LLM only responsible for formulating deep insights from the agents into a cohesive form that the human user easily understands.

Raina explains that Hero’s recipe for deriving intelligent insights is two-fold: data and intelligent AI agents operating on that data. Typical examples of the type of AI-driven insights offered by Hero include the flow of a specific token from CEX/DEX on the blockchain; the health of a token based on its founders, headquarters, and even its tokenomics; the token sentiment from Telegram groups and how correlated the sentiment has been in determining whether a token is bullish/bearish; analysis from popular indicators like MACD, RSI and Bollinger Bands.

Apta, which powers Hero AI Search, is based on the complex interaction between crypto-specific AI agents. One of these agents is specifically a rug pull detector. “Unfortunately, rug pulls are common in the crypto space. Fortunately, with access to a wealth of data, it is possible to extract trends that can be used to flag rug pull detectors,” Raina states.

The rug pull detector AI agent is trained on historic metadata to identify the probability of a given token being a scam. This agent can then directly be applied to predict the risk of fraud when new tokens are launched.

AI-based technologies are already beginning to impact the Web3 and DeFi space. However, Raina sees a major challenge to this exponential growth in the use of AI technology: compute. “All AI systems need appropriate computing infrastructure to run the systems efficiently. With the trend of ever-increasing AI model sizes, there is a strain on the hardware that hosts billions of parameters. Therefore, like many other domains, Web3 technologies need to explore opportunities with smaller AI models while still benefiting from the capabilities of the mammoth AI systems. The solution is distilling these large models onto smaller systems tuned for acute tasks required for specific Web3 technologies,” he notes.

However, when it comes to AI and finances, Vatsal Raina makes clear that before looking to use any AI tool for financial decision-making, it is paramount to understand what the AI is designed to do. Once that understanding is clear, inappropriate use of the technology is easier to avoid. He explains that a simple example of misuse of AI is that when ChatGPT was first launched by OpenAI in 2022, many users began to ask it complex mathematical questions. However, these users failed to appreciate that ChatGPT was designed to be a language model, not a calculator. “My advice is simple: any AI tool must only be used for what it has been designed to do for the best results,” Raina concludes. !

Link to visuals: https://drive.google.com/drive/folders/1EYo56r8CXYOYWI1MTpAnKs2OoVtfqMnp?usp=sharing