The emergence of Web3 signifies a transformative shift toward a decentralized internet that prioritizes user sovereignty, credible neutrality, and permissionless compute. However, decentralized blockchain applications are still primitive compared to their feature-rich TradFi and Web2 counterparts, primarily due to compute limitations as well as lack of tooling for advanced dApp development. This can be highlighted by the fact that expressive forms of compute leveraging powerful AI/ML-driven workflows are ubiquitous in traditional software applications whereas the integration of such technologies in crypto apps is a relatively underexplored design space. As the crypto ecosystem grows and evolves, integrating AI/ML is increasingly becoming more important in order for these applications to enhance features, functionality, and user experience to ultimately reach parity with Web2 applications.
In this blogpost we outline 10 real and impactful verticals of applied AI/ML for Web3 applications, that we’re either building at OpenGradient or working with one of our early launch partners on. The ten verticals are: protocol optimization, portfolio management, reputation systems, business intelligence and analytics, AI agents, risk management, natural language interfaces, prediction markets, AI-generated content, and smart contract security / MEV.
Note: There are many more verticals in the web3 application space that aren’t covered here, but we’re starting with ten that we’re actively tackling and are excited about.
1. Protocol Optimization
Leveraging artificial intelligence and machine learning for protocol or application optimization has become ubiquitous in the Web2 world; ranging from feed recommendation algorithms for social media to targeted advertising on webpages, most Web2 apps leverage AI/ML to deliver an improved user experience. Similarly, the same technologies can analyze vast amounts of historical data to adjust parameters in blockchain protocols and applications to deliver a superior and more optimized user experience.
Protocol settings like liquidity pool parameters, asset pricing curves, slippage limits, and trading fees in AMMs, lending protocols, vaults protocols, or yield aggregators can be optimized based on forward-looking forecasts on market movements and order flow. Systematic parameter hyperoptimization can help protocols protect user liquidity while generating better returns without slow and inefficient governance procedures. These types of secure and seamless integrations of intelligent workflows can not only improve the technical performance of these protocols but also enhance user experience by establishing a more stable and efficient ecosystem. Outside of DeFi, AI and ML models can also be leveraged for use-cases like dynamic gas pricing for network transactions, graph-based recommendation algorithms in SocialFi, and more.
One of OpenGradient’s first areas of research was how machine learning models can be leveraged to optimize AMM trading fees dynamically in order to help reduce LVR from arbitrageur flow and ultimately increase fee collection for LPs and reduce net impermanent loss. Supported by the Uniswap Foundation through a research grant, we published an extensive study here that shows how our initial set of models that scales fees dynamically based on volatility forecasts could increase LP fee collection for asset pairs like BTC/USDC by up to 18% in our historical simulations. In collaboration with other research and protocol partners like VectorDex who are productionalizing these research initiatives, the OpenGradient team continues to research how AI/ML can be leveraged to optimize a wide variety of different Web3 protocols to help ultimately improve user experience.
2. Portfolio Management
Portfolio management plays a critical role in modern finance by helping investors balance risk and return while working toward their financial goals. Modern portfolio management techniques can help traders systematically evaluate the return and risk profiles of thousands of different assets, allowing them to optimize portfolio allocations to maximize risk-adjusted returns. These types of analytical practices have become a cornerstone of traditional finance that touches virtually every aspect of the industry, from individual retirement accounts and mutual funds to massive institutional investors like hedge funds and sovereign wealth funds.
We have been exploring the application of these practices in crypto both internally and in partnership with its other clients, building spot forecasting models, volatility forecasting models, and more. These areas of research and the ML models developed can be very useful in creating traditional Modern Portfolio Theory (MPT) portfolio management workflows that leverage mean-variance analysis to construct efficient-frontier portfolios that optimize for metrics like Sharpe ratio. OpenGradient’s infrastructure and models help developers leverage these primitives to create powerful ML workflows and apply them to DeFi to allow protocols, apps, or users to systematically manage their assets more effectively. Volatility and risk models on OpenGradient are already being actively leveraged by launch partners like RNDM to help verifiably maximize risk-adjusted DeFi yields on their protocol.
3. Reputation Systems
AI/ML models can be leveraged to create dynamic, multi-dimensional scoring mechanisms for reputation systems that go beyond simple metrics like uptime, liveness, or response time. These systems can process complex combinations of on-chain and off-chain data, including network performance metrics, user feedback, historical reliability, geographic distribution of nodes, and cross-network participation patterns to generate comprehensive reliability scores. AI models, particularly deep neural networks and graph-based models, can also be used to detect subtle patterns of malicious behavior, identify collusion between nodes, and predict potential service degradation before it impacts users.
OpenGradient is actively collaborating with Reppo Labs, a spinout from Protocol Labs, to build AI-driven reputation systems and models for DePIN networks. As part of this initiative, OpenGradient is currently hosting many models that Reppo Labs and the Filecoin Foundation have created, including a Miner Performance Prediction Model and a Filecoin Allocator Reputation Model that are designed to use operation metrics as features for the models to dynamically score the performance of nodes.
This can be extremely impactful for networks that have critical reliability dependencies on node performance and operation, like DePINs. By leveraging AI-driven dynamic reputation systems that cryptoeconomically incentivize good behavior from nodes, peer-to-peer network reliability and trustworthiness guarantees can be improved over time. The OpenGradient team is excited to continue pushing on areas of research where applied modeling can be used to optimize infrastructure, and is also actively collaborating with teams like Peri Labs to explore reputation systems and other applied modeling use-cases for DePINs.
4. Business Intelligence and Analytics
Statistical analysis and business intelligence are important tools for understanding and optimizing software applications across both traditional web platforms and blockchain-based systems. In Web2 environments, analytical tools track user behavior, measure conversion rates, and monitor system performance, enabling teams to make data-driven decisions about feature development, UX improvements, and infrastructure scaling. In Web3, statistical analytics take on additional dimensions by incorporating on-chain data. Teams can analyze transaction patterns, smart contract interactions, and trades to understand how users engage with decentralized protocols.
Advanced analytics can be especially valuable for DeFi applications, where complex financial interactions generate massive amounts of data that can be analyzed to provide deeper insights. Machine learning algorithms can detect anomalous trading patterns that might indicate market manipulation or security breaches, while predictive models can help assess the health of lending pools and liquidity provision to evaluate systemic risk in the event of large market moves. Furthermore, network analysis can map relationships between addresses and contracts, providing insights into protocol usage and user behavior patterns to flag malicious wallets or sybil addresses in an otherwise opaque environment. Analytical capabilities for smart contracts that can provide verifiable on-the-fly intelligence is crucial for not just providing deeper understanding of user behavior, but also allowing developers to leverage those insights to, in turn, build more robust and efficient Web3 applications that better serve their users.
OpenGradient’s infrastructure allows developers to leverage powerful models in their smart contracts to build powerful analytical applications that provide business intelligence on the fly in a secure and verifiable fashion. In addition to in-house models we’ve developed, we also host models developed by partners like Pond, who has developed a powerful ML model to classify and flag Sybil addresses. Sybil address detection can be a powerful tool for running equitable airdrops, preventing referral system or incentive program abuse, flagging wash trading swarms, and providing more accurate user metrics.
5. AI Agents
AI agents are emerging as a significant industry vertical in both Web2 and Web3, serving multiple functions that enhance automation and decision-making capabilities. Unlike traditional software bots that follow a fixed heuristic, AI agents are more sophisticated and can understand context, adapt to new situations, and make complex decisions enabled by LLMs and agentic reasoning frameworks.
Blockchain technology offers unique advantages for AI agents, particularly in their ability to interact with decentralized networks. Beyond eliminating traditional identity verification requirements for financial operations, the open, permissionless, and programmable architecture of modern blockchains enables AI agents to seamlessly interact with application smart contracts, other AI agents, and data sources through standardized interfaces. AI agents in crypto has come to emerge as an important vertical that deserves its own blogpost, as a start Teng-Yan has a great write-up that outlines how AI agents will be a catalyst for crypto adoption as well as how the future of AI agents will be driven by coordination of specialized AI agents.
OpenGradient is working with many launch partners on building powerful agents that can augment different respective Web3 applications. One of them is Sparsity, a high-performance blockchain designed for high-throughput applications that is leveraging OpenGradient’s inference and reasoning infrastructure to build dynamic AI agents for their fully on-chain games.
6. Risk Management
Risk management in finance is crucial as it helps organizations and investors protect their assets and maintain financial stability by identifying, assessing, and mitigating potential threats to their portfolios and operations. AI and ML models that can be extremely effective at learning and extracting patterns from past data to help evaluate and mitigate sources of credit risk, fraud risk, inventory risk, idiosyncratic and systemic market risk. Advanced anomaly detection systems powered by ML can also safeguard against potential exploits or unusual trading patterns that might indicate increased risk for rugpulls or market manipulation, thereby protecting both protocol and its users.
OpenGradent is exploring various avenues of risk, including current research in optimizing LTV ratios in lending protocols using ML-driven risk models. Most current protocols use static parameters and employ little to no risk management practices, which could be significantly improved for the benefit of both users and the protocols through dynamic models that adjust collateral requirements and triggers depending on expected risk determined via ML models. Proper risk management for protocols can also help make DeFi much more palatable for institutional investors or asset managers.
7. Natural Language Interface
Natural language interfaces represent a significant leap forward in making blockchain technology more accessible to mainstream users by abstracting away the technical complexities that often create barriers to entry. Instead of requiring users to understand cryptocurrency addresses, smart contract functions, or transaction parameters, natural language interfaces allow them to interact with blockchain applications through simple, natural language commands on a ChatGPT-esque terminal. For example, in DeFi, users could describe their investment goals and risk tolerance in plain English, receiving recommendations for appropriate actions to take on different protocols. In NFT marketplaces, collectors could use natural queries to find specific artwork styles or track creator activity. By combining natural language processing with smart contract automation, these interfaces can also help users understand and verify what actions they're taking, reducing the risk of costly mistakes and making blockchain technology easier to use for new users.
In addition to providing superior user interfaces, language models can also enable net-new features that help augment user experience in Web3. Projects like Index Network are not just using OpenGradient’s SDK to leverage hosted LLMs for AI agents, but they’re also leveraging embeddings models for on-chain indexing and semantic search to create a graph-based discovery protocol that uses personalized recommendations to help users uncover new users, experiences, and applications.
8. Prediction Markets
Prediction markets in crypto can utilize AI and ML models to analyze vast amounts of historical trading data, on-chain metrics, social sentiment, and market indicators to generate more accurate probability estimates. Additionally LLMs can also be used to provide resolutions to markets instead of relying on human-based voting protocols; for example, Chaos Labs is working with Wintermute to create prediction markets that leverage “AI oracles” that can help provide accurate feeds for risk evaluation and tamper-proof resolutions.
Furthermore, machine learning model forecasts can be leveraged to price prediction markets and provide market-making liquidity for speculators. Double-Up is an on-chain gaming platform leveraging OpenGradient’s trust-minimized AI workflows and spot forecasting models to set prediction market moneylines to allow users to bet on returns of tokens like $SUI, and is the first of its kind that will be launching in December 2024.
9. AI Generated Content (AIGC)
AI can contribute to content generation by creating digital assets like text, images, and multimedia. Diffusion models can produce original content for decentralized platforms or projects, including art for NFTs; similarly, LLMs can generate text that can be used for a variety of other virtual assets. One such project is Sekai, which leverages a variety of AI models to allow creators to utilize tokenized on-chain IP assets to spin up their own visual novels in a matter of minutes.
Many projects also leverage decentralization as a means to source more GPU compute for less cost for AI inference. Projects like Aethir and Hyperbolic are creating networks that aggregate GPUs as a means to offer more cost-effective solutions to hosting and inferencing AI models, which can be a great solution amidst a global GPU shortage, sky-high cloud compute costs, and increasing centralization of AI technology. OpenGradient similarly leverages a completely decentralized filestore and a blockchain network to create infrastructure enabling credibly neutral and censorship-resistant access to AI models and workflows at all times, powered by a web portal that makes this technology accessible.
10. Security / Value Extraction
MEV involves extracting value from transaction ordering. AI can determine profitable transaction sequences for miners or validators while considering network health. Optimizing transaction sequencing helps capitalize on opportunities without compromising integrity.
ML models can detect and prevent front-running by identifying suspicious patterns and adjusting transaction propagation methods. AI may also assist in developing MEV strategies that minimize negative impacts like increased gas fees, contributing to a healthier ecosystem.
Mamori is one such company that has led research leveraging stochastic machine learning techniques like particle swarm optimization (PSO) to search for the maximum exploitable value given a particular state. Dynamic path-finding with ML offers significant advantages over more traditional methodologies like static analysis and can be more comprehensive in discovering more complex multi-block exploits while thoroughly examining and testing an input space for vulnerabilities.
Additionally, companies like Pond focused on developing foundational models for crypto use-cases have also dived deep into researching how deep learning with graph neural networks can be used to enhance security solutions for Web3 users. By observing the same behavioral patterns often exhibited by malicious contracts and wallets, graph neural networks can be great at learning from the graph-structure of on-chain transactions and subsequently leveraging frequently-observed patterns to detect and classify potentially suspicious or malicious actors.
Conclusion
As Web3 continues to mature, the integration of AI will likely become not just advantageous but essential for protocols and applications aiming to compete with traditional Web2 applications. The ability to leverage machine learning for enhanced decision-making, improved user experiences, and robust security measures will be crucial in bridging the current functionality gap between Web3 and traditional applications. Moreover, the unique combination of AI's analytical capabilities with blockchain's transparency and trustlessness creates opportunities for innovations that weren't possible in either domain alone.
Looking ahead, we can expect to see even more sophisticated applications emerge as these technologies continue to evolve and converge. The groundwork being laid today in areas like AI agents, dynamic reputation systems, and intelligent protocol optimization will likely form the foundation for a new generation of Web3 applications that are more efficient, secure, and user-friendly than ever before. The future of Web3 is not just decentralized – it's intelligent, adaptive, and increasingly sophisticated, powered by the transformative capabilities of artificial intelligence.
About OpenGradient
OpenGradient is a leading decentralized AI platform for open-source model hosting, secure inference, agentic reasoning, and application deployment. By developing tooling and a feature-rich platform that makes developing AI workflows both secure and seamless, OpenGradient empowers developers with the ability to build in our ecosystem of intelligent and optimized AI-empowered applications. With native model hosting, permissionless composability, and secure inference execution, OpenGradient also aims to accelerate open-source AI by democratizing model ownership, improving verifiability guarantees, and promoting censorship-resistant model access.
Current live products and initiatives include:
Research - In-house research team focused on developing AI and ML models as open-sourced public goods for Web3 protocols.
Model Hub - A Web3 model repository like HuggingFace, featuring a web UI built on a completely decentralized backend including OpenGradient’s decentralized filestore and blockchain infrastructure.
OpenGradient SDK - The OpenGradient SDK allows programmatic access to our model filestore and decentralized AI infrastructure from Python or from their CLI. Model developers can use the SDK to publish and manage their model deployments on OpenGradient and integrate it directly into their AI workflows.
NeuroML - Solidity library used to expose functions that are important for running AI-driven workloads on our EVM network, e.g. Data preprocessing, data post-processing, statistical analysis, different flavors of inference…etc.
Blockchain EVM Network - Blockchain Network leveraging heterogeneous AI compute architecture (HACA) and node specialization to support features like model hosting, data access, and AI inference secure and scalable.