Definition
CoreRec is the application of deep learning to recommendation systems that establishes an elegant, unified framework for research and production.
Simple by design.
A unified API for every stage of recommendation. Load data, pick a model, and train. Whether it's Matrix Factorization or Deep Learning, the code looks the same.
from corerec.engines.dcn import DCN
from cr_learn import ml_1m
from sklearn.model_selection import train_test_split
data = ml_1m.load()
train_df, _ = train_test_split(data['ratings'], test_size=0.2, random_state=42)
model = DCN(embedding_dim=64, epochs=20)
model.fit(
user_ids=train_df['user_id'].values,
item_ids=train_df['movie_id'].values,
ratings=train_df['rating'].values,
)
model.save('artifacts/my_dcn')
recs = model.recommend(user_id=1, top_k=10)
Retrieval at Billion-Scale.
Alibaba-style Tree-based Deep Models (TDM) break the vector search limit. Retrieve from billion-item catalogs in O(log N) time with beam search deep networks.
Neural Architecture Search.
Let CoreRec design your model. Reinforcement learning automatically discovers optimal neural architectures that balance accuracy, latency, and model size.
Optimize Clicks & Conversions.
Why choose one metric? Train multi-task models like Google's MMoE and Tencent's PLE to balance competing objectives—maximizing engagement while minimizing churn.
Deep Multimodal Fusion.
Go beyond collaborative filtering. CoreRec fuses text, image, audio, and behavioral embeddings using attention mechanisms to understand content.
Production Ready Serving.
Built-in serving layer handles real-time inference with minimal latency. Deploy your trained models to Kubernetes or bare metal environments.
Engine Architecture
Built on PyTorch. Optimized for performance. Ready for production.