The problem TokenSpector solves

Problem Statement:

In the burgeoning field of cryptocurrencies, transactions can often be fraught with risks. Fraudulent addresses posing as legitimate entities are not uncommon, leading to phishing scams that can result in substantial losses for unsuspecting individuals.

Solution: TokenSpector

TokenSpector is a chrome extension designed to assuage these concerns. It identifies potential phishing or scam Ethereum addresses based on their transaction data. It leverages a Custom Fine-Tuned GPT Model trained on the dataset of scam/phishing addresses.

Detailed Analysis:

TokenSpector employs a comprehensive set of over 25 transaction metrics to perform its analysis, including:

1. The average time between sent transactions
2. The average time between received transactions
3. Time difference between the first and last transaction
4. Number of sent transactions
5. Number of received transactions
6. Number of created contracts
7. Unique addresses from which Ether has been received
8. Unique addresses to which Ether has been sent
9. Minimum, maximum, and average values of received Ether
10. Minimum, maximum, and average values of sent Ether
11. Total Ether sent and received, and the resulting balance
12. Unique addresses involved in ERC20 token transactions
13. The average time between ERC20 token transactions
14. Minimum, maximum, and average values of received and sent ERC20 tokens

These metrics, among others, form the basis of TokenSpector's robust analytical framework.

User-Friendly:

TokenSpector is built with a user-friendly approach in mind, making it easy for individuals, regardless of their technical expertise, to verify the legitimacy of Ethereum addresses.

Challenges I ran into

Data Colelction:

Collecting and preparing the dataset was one of the most challenging aspects of this project. I started by scraping data from cryptoscamdb.org, which was a comprehensive but complex task due to the depth and variation of data.

Designing Metrics

Once the data was collected, I had to calculate 25+ metrics for each entry.

Coming up with relevant metrics to determine the issue was a challenging task in itself. It involved a deep understanding of the Ethereum network, scam patterns, and transaction data.

Fine tuning GPT-3

Fine-tuning the GPT-3 model with the custom dataset was another significant challenge. I had to prepare the prompts carefully to be fed into the model to give the most accurate output.

Building the extension

Moreover, building a chrome extension was something I wasn't very familiar with. Learning and implementing this in a short span of 24 hours was indeed a very challenging task.

Limited Time

It was also a challenge to balance precision and speed, as the chrome extension had to deliver accurate results quickly to ensure a good user experience. Additionally, handling errors and edge cases, such as when a user enters an address that isn't on the Ethereum network, was a tricky part of the project.

Beyond the technical aspects, managing the project within the limited timeframe was a significant challenge. The whole process, from idea conception, data gathering, model training, to application development, needed to be completed in a 24-hour window. This required careful time management and prioritization.

Despite all these challenges, working on TokenSpector has been a rewarding and enriching experience.