Cryptolytic is a platform for beginners tinkering with cryptocurrency to the seasoned trader. It provides you with recommendations on when to buy and sell based on technical indicators and assesses the markets to predict arbitrage opportunities before they even happen.
Our data science team:
- Utilized historical cryptocurrency data and created Random Forest models that predict upcoming arbitrage opportunities and crypto price movements.
- Built an end-to-end machine learning application on AWS using Lambda functions and an RDS PostgreSQL database to store data from the cryptowat.ch API, Sagemaker to train, S3 buckets to store, more Lambda functions make predictions with Random Forest models, and a Flask API deployed on Elastic Beanstalk to return current predictions.
How it works
We first gather historical data from each exchange’s API into a database for all of our supported exchanges and trading pairs and implement lambda functions in cloud9 to collect live data from the cryptowat.ch API in the respective tables. This allows us to have the most up to date data in our databases for predictions.
Random forest classifier models were trained on that dataset, stored in S3 buckets, and more Lambda functions were used to load those models and make predictions on the live data every 3 minutes. Those predictions were then inserted into the database in a new table that stored all of the predictions.
The Flask app retrieves the most recent predictions from the database and can return them to a user or be available for a backend team to use via API. The API was deployed on Elastic Beanstalk.
Overall we created a data pipeline and a backend that makes use of 30 different models to generate and store predictions on a recurring basis, which can then be accessed via API endpoints.
Role: App Architect, Data Engineer, Machine Learning Engineer, UI/UX Designer
DS Tech Stack: Python, SQL, Flask, Scikit, AWS (Elastic Beanstalk, RDS, Lambda Functions, Cloud9, KMS, Sagemaker), PostgreSQL
Timeline: 8 weeks
Our GitHub repo with all of the project code can be found here.