Your ML research feed,
ranked by your own AI
Personalized paper recommendations from ArXiv, a knowledge base you can ask questions to, and a learning-to-rank model that trains on your behavior.
What you get
Built for ML researchers
Personalized Daily Feed
ML-ranked papers from ArXiv and tech articles from Hacker News — tailored to your research focus.
Learns From You
Your own LightGBM model trains on your saves and dismissals. The more you interact, the smarter it gets.
Q&A Over Your Documents
Upload papers and notes. Ask questions and get cited answers backed by your personal knowledge base.
How it works
Simple, powerful, and personal
Set your interests
Choose research areas like NLP, CV, or RL. The system builds a semantic profile from your selections.
Generate your feed
ChromaDB HNSW retrieval finds semantically relevant papers across 20k+ indexed ArXiv papers in milliseconds.
Save and dismiss
Your interactions train a personal LightGBM ranking model. After 50 interactions, your model kicks in.
Live preview
Your daily feed
Papers and articles ranked by your personal ML model
Attention Is All You Need
ArXiv · cs.LG
·Score: 0.941
Introduces the Transformer — replacing recurrence entirely with self-attention for sequence modeling.
Transformers are now the backbone of every major LLM including GPT and BERT.
Directly relevant to your interest in NLP and deep learning architectures.
LightGBM: A Highly Efficient Gradient Boosting Decision Tree
ArXiv · cs.LG
·Score: 0.887
Histogram-based algorithms that dramatically speed up gradient boosting with lower memory.
LightGBM is the go-to model for tabular data and learning-to-rank in production ML.
Matches your interests in efficient ML and recommendation systems.
How Retrieval-Augmented Generation Actually Works
Hacker News · 342 points
·Score: 0.863
Covers the full RAG pipeline from document chunking to vector retrieval and LLM augmentation.
RAG is the standard approach for Q&A systems over private knowledge bases.
Highly relevant to your work on knowledge base assistants.
↑ Sample content — your real feed is ranked by your trained model and semantic interests