Task

• Pick 5 NFT collections of your choice
• Collect social media data and build metric of your choice that could be used in a pricing model
• The test should not take more than 2 hours, anything you cannot do, please write bullet points to explain your next steps Extra:
• questions to clarify assumptions
• design recommendations
• ideas of metrics/features for pricing model

#Schemas

OpenSea

Asset

id, integer
token_id , string
name, string
last_sale_total_price, string/null
last_sale_event_type, string/null
last_sale_auction_type, string/null
last_sale_event_timestamp, string/null
last_sale_quantity, int/null
asset_contract_address, string
asset_contract_created_date, string
asset_contract_name, string
asset_contract_schema_name, string
collection_creation_date, string/null
collection_discord_url, string
collection_name, string
collection_slug, string
owner_address, string

This resource is created by Ingestion.java and result can be found in spark-warehouse/ingestion folder. Unfortunately I still didn't receive my apikey for opensea and requests are failing due to rate limiter. I was able to pull data from one collection only.

Collection

created_date, string
description, string
discord_url, string
external_url, string
name, string
slug, string
stats_count, int
stats_num_owners, int
stats_average_price, float
stats_market_cap, float
stats_total_sales, int
stats_total_volume, float
primary_asset_contracts_address, string
primary_asset_contracts_symbol, string
primary_asset_contracts_opensea_version, string

Parts of this schema are ingested through ingestion process into Asset schema. If needed, we could pull more details from json responses.

LastSale

total_price, string
quantity, integer
event_timestamp, string
event_type, string
auction_type, string
successful, boolean

Some parts of LastSale were added to Asset schema through ingestion process. We could pull more details if needed.

Twitter

author_id, string
created_at, string
id, string
lang, string
text, string

Twitter search is done with SocialEnrichment.java and that was last thing I managed to deliver in 2 hours timeframe. I can also see that tweets search returned a lot of irrelevant bot activity tweets. That would need to be filtered down and query improved. Example of tweets csv should be in spark-warehouse/tweets folder.

TODO:

• list of retweet users
• list of like users Once we get the list of tweets above we should make another call to Twitter api to get more details about each tweet (likes, retweets). That should be fairly easy, but I run out of time.

Transaction

• token_id
• minted on OS or not
• collection name
• collection id
• list of transactions
• last price data
• list of auctions

TODO: Run out of time to implement transactions, the transactions could be pulled down from opensea (https://docs.opensea.io/reference/retrieving-orders), or alternatively we could use third party API that reads transactions from etherum network directly, because we know contract and id of token.

#Use cases

• auction price vs last transaction price
• average of last 3
• timelapse between last and last 3(closeness metric)

Insight into current state

• monax is building pipeline for the analytics that will power part of their product

Design

I created two spark applications, first will ingest data about nft tokens and second will pull data from twitter about corresponding nft collections. We could pull more data about transactions from opensea or other 3rp party api to be able to run som models on data. Outcome csv should be in spark-warehouse folders. Going to future, we would deploy this app as an AWS stepfunction that could be triggered by events. The AWS step function would consist of java spark jobs (either the 2 that I created in this demo: Ingestion.java& SocialEnrichment.java ) or we could create more jobs as we need. The spark would run in EMR cluster. EMR cluster is hadoop fleet of EC2 instances that work in distributed manner on our jobs and after the jobs are done, then the EMR cluster would be terminated. The output would be stored in s3 buckets as parquet files, so that we can do further processing on our internal data, or create own data warehouse, or we could create athena tables using AWS glue for analysis run by Scientists.

PS: There are just so many improvements that we can do with this, so the best would be to have a chat and answer your questions directly

Best regards, Stefan Kecskes