9 min read
The Transformer, One Attention Head at a Time
A ground-up walk through self-attention — queries, keys, values, and why scaled dot-product attention became the backbone of modern AI.
Explore by topic
Categories
Filter the archive by research area — from LLMs and fine-tuning to evals, embeddings, and inference infrastructure — and go straight to the depth you want.
Browse by category
Pick a topic to narrow the feed, or view every article across the archive.
8 min read
A Practical Guide to RAG Evaluation
Retrieval-augmented generation is easy to demo and hard to measure. Here is how to separate retrieval quality from generation quality.
8 min read
Fine-Tuning Without Melting Your GPU: A LoRA Deep Dive
Low-Rank Adaptation trains a fraction of the parameters and still adapts a large model. Here is the intuition and the code.
7 min read
Evals Are the Product: Building Benchmarks You Can Trust
Model quality is whatever your eval measures. If the benchmark is weak, the whole team optimizes the wrong thing.
8 min read
Embeddings and Vector Search, From Cosine to ANN
How text becomes geometry, why cosine similarity works, and how approximate nearest neighbor search makes it fast at scale.
9 min read
Serving LLMs in Production: An Inference Infrastructure Primer
Batching, KV caching, and quantization are what stand between a working model and an affordable one. A tour of the levers that matter.