BTC 3-Minute Direction Predictor
Budget: $30 – $250 USD
I want a lightweight application that ingests historical BTC price data, learns from it, and then tells me— with at least 70 % measured accuracy—whether the price will be higher or lower three to four minutes after any given moment. Once the prediction is made, the app must immediately push a simple “Up” or “Down” message to a messaging channel; I am flexible here, so you can wire it to Telegram, Discord, or both if that streamlines your build.
Accuracy is paramount, so the model should come with a back-test report that clearly shows how you calculated the 70 %+ success rate using unseen historical candles. A fast inference time is also important; the whole pipeline from data capture to message dispatch should stay comfortably under the four-minute window.
Deliverables I need to review and sign off on:
• Source code (Python is preferred, but a different language is fine if you justify it)
• Trained model and training notebook or script
• Back-testing results proving the stated accuracy
• Deployment instructions and a small script or bot that forwards the Up/Down signal to the chosen channel
Please include a detailed project proposal that walks me through your approach to feature engineering on historical data, the model architecture you favour (e.g. LSTM, gradient boosting, lightweight transformer, etc.), the libraries you’ll use (pandas, scikit-learn, PyTorch/TensorFlow, telethon / discord.py, etc.), and an estimated timeline for each milestone.
Accuracy is paramount, so the model should come with a back-test report that clearly shows how you calculated the 70 %+ success rate using unseen historical candles. A fast inference time is also important; the whole pipeline from data capture to message dispatch should stay comfortably under the four-minute window.
Deliverables I need to review and sign off on:
• Source code (Python is preferred, but a different language is fine if you justify it)
• Trained model and training notebook or script
• Back-testing results proving the stated accuracy
• Deployment instructions and a small script or bot that forwards the Up/Down signal to the chosen channel
Please include a detailed project proposal that walks me through your approach to feature engineering on historical data, the model architecture you favour (e.g. LSTM, gradient boosting, lightweight transformer, etc.), the libraries you’ll use (pandas, scikit-learn, PyTorch/TensorFlow, telethon / discord.py, etc.), and an estimated timeline for each milestone.