Modular AI Tech Stack: Market Landscape and Key Players in 2024

Artificial intelligence's (AI) rate of growth and adoption is quite unheard of in the history of tech. The accelerated efforts on improving the infrastructure, backend and developer side, coupled with user adoption at an extraordinary scale — are making it the youngest backbone of our digital infrastructure.

All this growth and immense potential brings a profound challenge: 

How do we ensure that these powerful algorithms remain accountable, transparent, and aligned with human interests?

Decentralized AI (DeAI) offers a solution as a modular tech stack distributing authority across multiple projects and protocols. Every component of the tech stack from data storage to model training, from computation to governance — is an independent building block within DeAI.

In this ecosystem guide, we'll explore the core components of decentralized AI tech stack, major projects in that space, and how all of them work together.

What is Decentralized AI Tech Stack?

Decentralized AI tech stack is a modular collection of technologies, tools, protocols, and projects needed to develop and deploy AI solutions. Cryptographic protocols for verifiable compute, decentralized storage, identity protocols, open-source inferencing models, and more come under the umbrella of a DeAI tech stack.

The promise of this tech stack is to solve three key problems that centralized AI has brought up:

1. Lack of transparency (what data is used and how?) and computational integrity (did the model run correctly?). 
2. Gatekeeping of AI infrastructure through permissioned APIs, hardware scarcity, and exorbitant custom finetuning costs.
3. Users have zero say in how their data is being collected and processed.  

DeAI promises a more open, democratic, and accountable AI infrastructure for developers and users alike. 

Now, let's see how each component of the tech stack contributes to delivering the DeAi promise.

Data Storage

Decentralized AI systems require verifiable storage solutions to handle the massive datasets of training models, raw data, and managing inference results.

Two major projects in the space are Filecoin and Arweave. 

Filecoin

Filecoin transforms cloud storage into a decentralized, algorithmic marketplace, allowing users to rent out unused storage space and monetize it securely. In the context of DeAI, Filecoin’s marketplace can serve a few use cases like:

• Proof-of-replication validation for ensuring data integrity in training sets.
• Integration with CDNs for low-latency data delivery to inference endpoints.
• Cross-network data availability through libp2p networking and content-addressed storage.

Arweave

Arweave offers a decentralized storage solution focused on permanence, allowing data to be stored indefinitely through a unique ‘blockweave’ structure. 

Data permanence is a superpower that Arweave brings to DeAI, especially to those use cases/applications that require historical datasets and model versions to be preserved and kept accessible.

Potential use cases of Arweave with DeAI are:

• Permanent storage for model architectures and training configurations.
• Immutable audit trails for training processes and model lineage.
• Version control systems for model iterations.
• Permaweb-based distribution networks for AI model sharing.

Data Collection

Data collection is a key component of DeAI for two main purposes:

1. Ensuring DeAI, at core, is powered by high-quality data.
2. Empowering users to retain control over their data while contributing to DeAI i.e. potential for incentivization.

Currently, there are two two projects that are procuring high-quality training data for AI models in a decentralized way i.e. ethically sourced, incentivized, and privacy-preserving — Grass Protocol and VANA.

Grass Protocol

Grass Protocol is a data provisioning protocol, meaning it scrapes raw data from users and makes it available to AI companies. On a decentralized front, Grass incentivizes users for their data contribution, has a permanent metadata to prove the data's origin, and has several privacy measures in place.

There are so many possibilities for Grass Protocol within the DeAI tech stack:

• Cryptographic verification of data sources
• Incentivized data contribution mechanisms
• Privacy-preserving data collection pipelines
• Reputation systems for data providers
• Custom data collection workflows

Vana

Vana enables individuals to collect, own, and monetize their personal data, giving users control over its use while preserving privacy. Vana serves the need for diverse and user-consented data while offering users the ability to monetize their data. This data ownership model can empower DeAI with,

• Personal data vaults with granular access control
• Federated dataset creation tools
• Data monetization frameworks at a retail level
• User-controlled data permissions
• Proof-of-contribution mechanisms

Data Preparation

Data preparation in the DeAI life cycle involves indexing, cleansing, and organizing of data to make it ready for model training and inference.

Four key projects in the data preparation stage are: Tableland, Space and Time, Goldsky, and Noves.

Tableland

Tableland is a decentralized database protocol that transforms raw data into structured, queryable formats suitable for model training and inference.

By providing a robust, SQL-compatible structure for data management, Tableland enables DeAI projects to create, read, update, and delete data tables on-chain. 

Integrating Tableland with decentralized AI workflows allows for consistent, verifiable data access across different stages of AI development.

Space and Time

Space and Time serves as a decentralized data warehouse combining on-chain and off-chain data allowing developers to run tamper-proof queries. This capability is highly beneficial in DeAI, where models often require consistent, verifiable data across various sources. 

Within DeAI, Space and Time can enable more comprehensive data preparation like:

• Tamper-proof query execution for training data integrity.
• Ensure consistency across training and inference tasks.
• Real-time data aggregation from multiple sources.
• Proof of SQL execution for verifiable computations.

Goldsky

Goldsky provides real-time data streaming and indexing services tailored for blockchain applications. 

In the context of DeAI, Goldsky can help dApps (decentralized applications), DeFi (decentralized protocols), and other web3 products with,

• Constant stream of data, ranging from mere users transaction to state changes by smart contract execution.
• Predictive analytics or sentiment analysis based on live data inputs.

Noves

Noves specializes in data analytics and visualization tools designed for interpreting complex blockchain data.

In a DeAI stack, Noves can supercharge data preparation with,

• Real-time monitoring of data quality, trends, or patterns within datasets
• Automated data cleaning and preprocessing
• Custom analytics pipelines for specific AI models

Model Training and Finetuning

Model training and fine-tuning are two critical tasks in the DeAI workflow. These processes give AI models and solutions the power to learn from data in a custom manner and adapt to new tasks.

Traditional training infrastructures are often centralized, expensive, and limited in accessibility, posing barriers for independent developers and smaller organizations.

However, companies like Gensysn, Bacalhau, and Flock.io are enabling DeAI projects to train and fine-tune models in a decentralized manner.

Gensyn

Gensyn is a decentralized compute protocol that orchestrates scalable machine learning training across distributed nodes. By offering scalable access to compute resources without relying on centralized providers, Gensyn lowers the barriers to model training, allowing developers to utilize a distributed network at reduced costs.

For DeAI projects, Gensyn’s marketplace can serve as an ideal backend for computationally intensive tasks. Apart from this, Gensys opens up possibilities like:

• Distributed training orchestration with probabilistic proof-of-learning.
• Automated checkpoint management and model synchronization.
• Custom training pipeline deployment with versioning.

Bacalhau

Bacalhau enables decentralized computation close to where the data is stored, particularly within distributed storage networks like IPFS and Filecoin. This is especially relevant for DeAI, where data often resides in distributed storage networks, and moving it for processing can be an efficiency or privacy compromise.

By integrating Bacalhau, DeAI projects can,

• Execute training and fine-tuning tasks near the data source.
• Have custom runtime environments for specific model architectures.
• Distribute workload across nodes with resource optimization.

Flock.io

Flock.io specializes in federated learning, a method that allows multiple entities to collaboratively train models on shared datasets while keeping data localized and private. This learning framework enables DeAI projects to,

• Build hyper-personalized and specific AI models or agents.
• Improve model accuracy using a broad range of data sources.

Compute and Inference

Compute and inference involves the application of trained models to process and analyze data, make predictions, and generate insights. This layer enables efficient model execution and scalable inference for the DeAI stack. Projects to highlight in this layer are Akash Network, ioNet, Modulus, and Ritual. 

Akash Network

Akash Network is a decentralized cloud computing marketplace that connects users with compute resources. It offers developers and organizations a scalable, secure, and on-demand access to compute infrastructure. 

This can help developers run AI inference tasks at scale using underutilized data center capacity worldwide. Some potential use cases arising out of this workflow are,

• Dynamic provisioning of compute resources. 
• Edge computing support for real-time inference.
• Custom deployment environments for specific models.

ioNet

ioNet is a decentralized compute network designed specifically to handle AI workloads. For DeAI projects, ioNet’s architecture offers an ideal environment for running intensive inference tasks, as its infrastructure is optimized to support high-throughput and low-latency processing. 

By utilizing ioNet, DeAI applications can build,

• Specialized hardware acceleration for AI workloads.
• Real-time inference optimization.
• Distributed model serving architecture.
• Cross-chain compute verification.
• Custom runtime environments.

Modulus

Modulus offers a decentralized platform for deploying and managing machine learning models across a distributed network of compute providers. Within a DeAI framework, Modulus could serve as the resilient backend for AI applications. It can offer superpowers like:

• ZK (zero-knowledge) powered verification of AI processes and outputs.
• Private identity authentication.
• Auto-scaling of AI computation resources.

Ritual

Ritual is another decentralized compute network focused on AI inference. It offers developers access to distributed model computations across a scalable set of nodes.

For DeAI projects, Ritual’s infrastructure means,

• Low-latency computation and inference optimization
• Custom runtime environments
• Edge deployment support

Actively Validated Services (AVS)

AVS provides a decentralized security layer that continuously validates the operations of networks and services. For DeAI, AVS can bring in computational integrity which builds trust and improves resilience. And the most prominent AVS provider is EigenLayer. 

EigenLayer

EigenLayer is a restaking protocol that allows Ethereum validators and stakers to re-stake assets to provide security and validation for new services. 

In DeAI, EigenLayer’s AVS can be integrated for,

• AI model validation and verification systems.
• Automated slashing mechanisms.
• Proof-of-computation verification.
• Validator reputation systems.
• Cross-chain security coordination.

Decentralized Identity

Decentralized identity enables users and AI agents to manage their credentials securely across AI applications, maintaining control over private information. Two major players working on decentralized identity (DiD) are ENS — Ethereum Name Service and Privado. 

ENS — Ethereum Name Service

ENS is a decentralized naming system on the Ethereum blockchain that maps human-readable names (e.g., "alice.eth") to complex addresses. In DeAI, ENS can streamline interactions by allowing users and AI agents to access models, datasets, and services using simplified identifiers.

The potential of ENS within DeAI can be extended to,

• AI model registry and discovery.
• Metadata management systems.
• Identity verification frameworks.
• Reputation aggregation.
• Access control management.

Privado

Privado facilitates trusted and secure relationships between apps and users. For DeAI, Privado can serve as the foundation for privacy-preserving identity verification for both users and AI agents. By integrating Privado, DeAI applications can verify credentials and maintain agent authenticity.

Not just that, Privado can bring in privacy and security measures to DeAI like,

• Enterprise authentication systems.
• Bot authenticity verification.
• Privacy-preserving credential issuance.

Confidential Compute

Confidential computing ensures that data remains private and protected during processing, which is essential for sensitive computations in DeAI. This is a must-have for DeAI to spread its use cases across fields like healthcare, personal banking, hardware security, and even government applications.

Two players that are working in this space are Jiritsu and LibertAI.

Jiritsu

Jiritsu offers confidential computing through Trusted Execution Environments (TEEs) and cryptographic techniques to process data securely and privately. 

Apart from the core thesis of allowing AI models to work with private and regulatory-sensitive data, Jiritsu also can potentially offer,

• Private inference execution
• Data privacy guarantees
• Encrypted computation pipelines
• Verifiable compute proofs
• Privacy-preserving analytics

LibertAi

LibertAI focuses on privacy-preserving AI by providing tools for confidential model training and inference. It employs multi-party computation and differential privacy to allow models to learn from sensitive data without exposing it.

From an integration standpoint, LibertAI comes with a ton of potential to contribute to the DeAI space:

• Private model training frameworks
• Data anonymization techniques
• Privacy-preserving aggregation
• Encrypted model sharing

Intent Solvers

Intent solvers optimize the execution of tasks and actions in decentralized networks, ensuring that user requests are processed efficiently and in an orderly manner. 

For DeAI, intent solvers can prioritize and manage AI-related tasks, enabling smooth coordination across distributed systems and reducing transaction costs.

A major project working on intent solver coordination is Epoch Protocol.

Epoch Protocol

Epoch Protocol is a coordination layer that aggregates and sequences user intents across decentralized networks. Integrations with Epoch Protocol can enable DeAI apps to handle high volumes of user requests or inference calls in a cost-effective and orderly manner. 

Apart from that, Epoch Protocol can bring about novel use cases like:

• Smart order routing for AI compute requests
• Real-time resource discovery and allocation
• Multi-chain compute orchestration
• Natural language intent parsing
• Network congestion monitoring 

AI Payments

AI payments layer facilitates the secure exchange of data, services, and computational resources. It enables efficient monetization of AI services, ensuring fair compensation for data providers and developers. Nevermined is the key protocol heading this space.

Nevermined

Nevermined is a decentralized AI payments protocol designed to facilitate seamless transactions between AI agents, effectively serving as the ‘PayPal of AI Commerce’. It provides the payment infrastructure for AI-to-AI interactions, enabling intelligent software agents to discover, negotiate, and compensate each other in real-time.

Additionally, Nevermined’s Web3 wallet integration allows for secure on-chain payments. This ensures that transactions between data providers, AI models, and users are transparent and verifiable.

Outside this, Nevermined can be a huge value boost to DeAI in monetization and payment verticals with use cases like:

• Automated settlement systems
• Agent-to-agent transactions
• Automated micropayments
• Token-gated service access
• Dynamic pricing optimization

Each individual component of the DeAI stack offers powerful abilities and each project promises a ton of solutions and use cases. However, this gives rise to the need for an effective orchestration layer which ideally should serve as the connective tissue between various DeAI components.

Gaia: The Orchestration Layer for Decentralized AI

Gaia is a comprehensive orchestration layer designed for DeAI and facilitates the interactions between different modules like data storage, identity protocols, and compute resources. With Gaia, each layer of the DeAI stack functions as part of a synchronized system. 

This allows developers to create, deploy, and govern AI agents without the friction of navigating siloed infrastructure or having to build a tech component from scratch.

By serving as an orchestration layer, Gaia powers DeAI with:

1. Automated compute resource allocation and load balancing across multiple chains and nodes.
2. Robust framework for deploying, managing, and securely operating custom AI agents.
3. An economic model that incentivizes participation, where node operators and contributors are rewarded through staking and revenue sharing.

Whether you're a developer looking to build custom AI agents, an enterprise seeking secure AI solutions, or a domain expert wanting to monetize your knowledge, Gaia provides the tools and infrastructure you need.

Here are some resources to help you get started with Gaia:

• Gaia User Guide: https://docs.gaianet.ai/category/gaianet-user-guide
• Chat UI: https://knowledge2.gaianet.network/chatbot-ui/index.html 
• Node installation: https://github.com/GaiaNet-AI/gaianet-node

How to install a Gaia Node:

or

How to Install a GaiaNet Node.pdf

Google Docs

Demo videos: