Unmarshal is Venturing into Retrieval-Augmented Generation (RAG): A New Era of Intelligent Blockchain Data Access

4 min readFeb 13, 2025

At Unmarshal, we’ve always been committed to empowering developers, enterprises, and blockchain enthusiasts with seamless access to indexed, real-time data from multiple blockchain networks. Today, we are thrilled to announce our next transformative step — Unmarshal is entering the world of Retrieval-Augmented Generation (RAG).

This marks a major evolution in how blockchain data is queried, processed, and delivered,

Why RAG?

The blockchain ecosystem is growing exponentially, and so is the complexity of on-chain data. Developers and users often face challenges in extracting actionable insights from this data. While Unmarshal APIs provide robust data indexing, an RAG-based approach takes it a step further by blending retrieval capabilities with the intelligence of generative AI models.

With RAG, we aim to:

Simplify Data Access: Empower users to query blockchain data in natural language.
Provide Context-Rich Responses: Augment raw on-chain data with AI-driven context and explanations.
Foster Smarter Decision-Making: Offer actionable insights that are relevant, accurate, and easy to understand.

What is Retrieval-Augmented Generation (RAG)?

RAG combines two powerful technologies:

Data Retrieval: The process of fetching relevant information from a database or knowledge source.
Generative AI Models: Large Language Models (LLMs) such as GPT or T5 that generate natural language responses based on input prompts.

In a RAG pipeline, real-time data from Unmarshal APIs will serve as the retrieval layer, while AI models will process this data to generate responses tailored to user queries.

How RAG Works at Unmarshal

Here’s a technical breakdown of how our RAG-powered pipeline will function:

1. Query Parsing and Interpretation

User queries are processed by a natural language understanding (NLU) module to identify intent, keywords, and required data fields.
Example: Query — “What are the last 10 transactions for wallet 0x123…?”

2. Data Retrieval

The parsed query is passed to Unmarshal’s multi-chain data indexing APIs.
Relevant endpoints (e.g., /wallet/transactions, /token/prices, /contract/events) fetch real-time on-chain data.
Retrieved data is then structured for augmentation.

Example API Response:
json
CopyEdit
{“timestamp”: “2025–01–01T12:00:00Z”,

“to”: “0x456…”,

“amount”: “1.5 ETH”,

“status”: “Success”}

3. Augmentation Layer

Retrieved data is fed into an augmentation layer that enriches it with additional context:
Historical trends (e.g., wallet activity over time).
Off-chain data (e.g., token market prices).
The output is structured into a prompt for the generative AI model.

Example Augmented Prompt:

CopyEdit

Query: “What are the last 10 transactions for wallet 0x123…?”

Data: [

{“timestamp”: “2025–01–01T12:00:00Z”, “to”: “0x456…”, “amount”: “1.5 ETH”, “status”: “Success”},

{“timestamp”: “2025–01–01T11:00:00Z”, “to”: “0x789…”, “amount”: “2.0 ETH”, “status”: “Failed”}

]

Augmented Prompt: “Provide a user-friendly summary of the latest transactions for wallet 0x123 based on this data…”

4. Generative AI Layer

The AI model (e.g., GPT or a custom fine-tuned LLM) processes the prompt and generates a human-readable response.
Example Response: “The latest transaction for wallet 0x123… occurred on January 1, 2025, at 12:00 PM, transferring 1.5 ETH to wallet 0x456… successfully. The previous transaction on the same day at 11:00 AM failed when attempting to send 2.0 ETH to wallet 0x789…”

5. Response Delivery

The final response is formatted and delivered via API or user-facing applications.

Infrastructure Overview

Data Layer:

Leverages Unmarshal APIs for real-time data retrieval across 45+ blockchain networks.
Includes decentralized storage options for indexing historical data.

Vector Database:

Implements scalable solutions like Pinecone or Milvus for efficient similarity search and retrieval of frequently queried data.

AI Layer:

Uses state-of-the-art LLMs, fine-tuned on blockchain-specific datasets to ensure domain relevance.

Integration Framework:

API gateway ensures seamless integration with existing dApps, enterprise solutions, and analytics platforms.

Use Cases Enhanced by RAG

DeFi Protocol Analytics: Real-time insights on token flows, liquidity, and performance.
NFT Market Intelligence: Contextualized data on NFT transactions, ownership history, and trends.
Smart Contract Debugging: Human-readable summaries of contract interactions.
Enterprise Reporting: Automated generation of compliance reports and blockchain audits.

Next Steps

This journey is just beginning. Over the coming months, we’ll:

Launch beta versions of RAG-powered APIs for developers to test.
Introduce open-source tools for building intelligent blockchain applications.
Partner with AI-first blockchain projects to expand the ecosystem.

Join the RAG Revolution

Unmarshal’s venture into RAG is about more than just technology — it’s about creating a smarter, more intuitive blockchain experience. Whether you’re a developer, enterprise, or enthusiast, we’re building tools that make blockchain data accessible, actionable, and intelligent.

Let’s shape the future of blockchain data access, together.

About Unmarshal

UnmarshalAI is a leading decentralized multi-chain data network, enabling Web3 projects to access accurate, real-time blockchain data across 50+ blockchains, including Ethereum, Base, Optimism, BNB Chain, Avalanche, Mode Network, and Fantom, making it one of the most versatile Web3 data providers.

Leveraging AI-driven solutions, Unmarshal enhances data accessibility and insights for RWA, DePIN, AI Agents, DeFi, NFT, and GameFi platforms. Through robust APIs, notification services, and no-code indexers, it empowers dApps to deliver seamless user experiences while ensuring transparency, scalability, and innovation at the forefront of Web3 advancements.