AI under the Hood (AUTH)
Instead of chasing the noise, we focus on the foundational concepts of AI. That is the mission of AI Under the Hood (AUTH).
AI is a multi-layer, full-stack system. While each layer is deep enough to define an entire career, this guide is designed to look at them as a single, connected architecture.
Our Approach: First Principles & Paved Paths
We follow a dual-track approach: we provide a paved path for you to follow, while enforcing first-principles reasoning at every step.
We acknowledge the limits of this approach. Since AI took center stage in late 2022, it has taken the world's finest minds over three years to effectively "tame the beast" through agentic systems. It would be ambitious to claim we can recreate every original idea from scratch in a tutorial.
However, we have the advantage of hindsight. By applying first-principles thinking to established breakthroughs, we give you the "why" behind the "how," ensuring you aren't just following a script, but building a mental model that lasts.
Note: This platform is a living resource and evolves at a rapid pace.
Read the blog posts or directly head ovet to the sample Code .
Recent Posts
-
Why Embeddings Matter
Updated:A deep dive into what embeddings are, why they matter, and how they power modern AI, semantic search, and RAG-based systems.
-
What LLMs Do at Inference: A Deep Dive Under the Hood
Updated:A step-by-step, reference-backed explanation of what happens during LLM inference: tokenization, embeddings, prefill & decode phases, KV caching, decoding strategies, bottlenecks and optimizations like quantization, FlashAttention and speculative decoding.
-
Transformers in AI
Updated:The Architecture That Revolutionized Machine Learning
-
Top-k vs. Nucleus Sampling
Updated:Decoding the Secrets of AI Text Generation