DeHive - Blaize

AI-driven asset management solution for DeHive

Smart Asset Management Solution based on Artificial Intelligence and Machine Learning for DeHive Protocol

Time-to-Market

6 months

Blockchain

Ethereum

Project Stage

Finished product

Overview

Asset management
Machine learning
Smart strategy

Case study illustrates, how will a decentralized project benefit from a separate AI/ML pipeline for effective asset management.

Blaize's reputation for delivering state-of-the-art blockchain solutions is further exemplified by its work on DeHive. Despite the inherent volatility of crypto assets, Blaize team managed to harness the power of AI/ML to introduce a smart strategy for asset management.

Task

The product for DeHive was developed by Blaize team internally. The blockchain team developed a system of decentralized indexes (clusters), which are derivatives pegged to a basket of crypto assets from a specific web3 area. A critical aspect of this project was the need for a separate AI/ML pipeline dedicated to rebalancing these indexes (clusters).

Given the unstable nature of crypto assets, managing index peg and price presented a non-trivial task. The solution was to devise an intelligent model capable of overseeing asset management. This model required a series of steps: data scrapping and cleaning; building the ML model; establishing a CI/CD pipeline for model retraining with new data; and finally, feeding the latest inputs into the model to determine confidence about internal index proportion changes and to provide recommendations on new parameters.

Technologies

Python
R
Machine Learning
Decision Trees
Logistic Regression
Non-linear Regression

Project Architecture

A dedicated AI/ML team worked with DeFi Analytics Team Lead to validate the model's output. This iterative development process involved multiple rounds of feedback and model fine-tuning.

The data scrapping team ensured the preparation of a clean dataset with the necessary input parameters. Then, the AI/ML team developed the model using machine learning techniques, including regression methods and decision trees. Once the initial model achieved successful outcomes, it was deployed to the pipeline. The team then focused on keeping the model up-to-date and expanding it to include new indexes added to the protocol.

Book a free call with Blaize experts to learn how to enhance your business using AI

Learn more

Q4 2022-Q1 2023
ML Pipeline Development

The data scraping team prepared the dataset with input parameters, including scripts for dataset updates and cleaning. Additional proprietary parameters were included in the model.

Q1 2023-Q2 2023
DeFi parameters development and data set preparation

The AI/ML team worked iteratively, collaborating with the lead of DeFi analytics team. Multiple combinations of solutions were probed and compared. Upon achieving the first successful model, it was integrated into the pipeline.

Q2 2023
ML Model deployment into production

The development process required cooperation between AI/ML engineers, data scrappers, and the DeFi analytics team. The AI/ML team improved the model, testing different techniques and parameters, while the DeFi analytics offered feedback to perfect the model.

Challenge

The development of the ML model posed two main challenges. Firstly, it was crucial to understand that the model is a manageable oracle, requiring the team to balance simplicity and effectiveness in its outputs.

Secondly, the iterative development of the model required testing various input parameters and technique combinations with a feedback loop for error correction. The expertise of the DeFi analytics Team Lead played a vital role in this feedback loop.

Result

Blaize's AI/ML team successfully developed a smart asset management model for DeHive. They built a pipeline that facilitated instant updates to the dataset used by the model, ensuring its accuracy and relevance. The intelligent strategy, powered by various ML techniques, analyzed assets behavior within the index and provided recommendations on proportion changes or maintenance. The model offered decision suggestions such as increasing or decreasing the value of specific assets in the index or removing them altogether.

These are crucial benefits of using and implementing an AI-driven solution for DeHive: