Are you constantly checking charts and wondering if it's the right moment to sell your precious Chainlink tokens? Do you obsessively monitor price movements, hoping for that elusive moonshot?

Well, your worries might just be over—Predict-A-Link has arrived!

Check Tomorrow's Prediction

Let me share the journey of building this project, which has been an incredible learning experience over the past four weeks.

My aim wasn't to create something perfect but to push myself beyond my usual boundaries and enhance my skills by embracing a new challenge. Committing to this project publicly kept me motivated to deliver a minimum viable product.

As someone who's always been fascinated by the intersection of technology and finance, I wanted to create a tool that could help others navigate the volatile crypto market more confidently.

Building Predict-A-Link truly pushed me out of my comfort zone, requiring me to tackle several significant challenges:

  • API Rate Limiting: Understanding and implementing proper API usage within allowed limits.
  • Predictive Modeling: Learning techniques for time series data analysis and forecasting.
  • Front-End Development: Exploring frameworks to create an intuitive user interface.

How Does the Backend Work?

The backend performs several crucial tasks to ensure daily predictions are displayed on the website. It works through three steps that seamlessly deliver daily price predictions:

Data Collection and API Retrieval

The system begins by gathering data from three crucial sources:

  • Crypto Price Data: Daily cryptocurrency price information gathered for Bitcoin, Ethereum, Sui, and Chainlink.
  • Search Trends Volume: Market sentiment and public interest metrics.
  • Financial Data: Relevant economic indicators.

This data is collected through a daily API retrieval process and is used to train the models, make predictions, and track the models' results.

Data Processing Pipeline

Once collected, the data flows through a processing pipeline that leverages SQL and Python:

  • The system stores all raw data in BigQuery.
  • Data undergoes preprocessing to ensure consistency and quality; additional features are created.
  • The processed information is prepared for model training and predictions, creating datasets for the different time horizon models.

Model Training and Prediction System

The heart of Predict-A-Link lies in its intelligent prediction system:

Weekly Training Cycle

The system springs into action every Monday, training fresh LSTM models using various input date ranges. This weekly retraining is essential as the Chainlink market is highly volatile, and the training will help our predictions stay current.

LSTM models are a type of recurrent neural network capable of learning order dependence in sequence prediction problems, making them ideal for time series forecasting.

We build five unique models, each based on its own set of dates. For example, the 13-day model makes predictions using data from the last 13 days. This weekly retraining ensures the models stay current with market trends.

Daily Prediction Flow

On non-Monday days, the system skips training and moves directly to making predictions. All predictions are systematically saved to BigQuery and Firestore (a flexible NoSQL cloud database) for reliable storage and quick retrieval.

This automated workflow ensures users can access fresh predictions while maintaining model accuracy through regular retraining cycles.

Building the Front End

The final piece of the puzzle was creating an interface to display these predictions—a part of the project that was entirely new to me. I focused on designing a clean, user-friendly experience that makes complex data easy to understand.

This journey involved extensive research, and leveraging Large Language Models like Claude and ChatGPT helped accelerate my learning. I used libraries such as Chart.js (an open-source JavaScript library for data visualization) to visualize the data and relied on Firebase to deploy and host both the web app and the predicted price data.

Final Words

Building Predict-A-Link has been an incredible four-week journey that allowed me to tackle new challenges in working with APIs, predictive modeling, and front-end development. From orchestrating the backend processes that seamlessly deliver daily price predictions to designing an interface that makes complex data accessible, every step was a valuable learning experience.

I'm excited to share this project and would love to hear your thoughts. If you're considering building in public, I can't recommend it enough—the commitment to sharing your progress can be a great motivator.

Feel free to try out Predict-A-Link and let me know what you think. 

Check Tomorrow's Prediction

Your feedback is invaluable and will help me improve the tool further.

Thank you for joining me on this adventure!