Interactive Paper Library

Research papers,
rebuilt to be explored.

Landmark AI papers as interactive articles — animated architectures, sortable results, three reading levels from beginner to researcher — with every equation and number preserved from the original.

LLM Complete Guide

2 of 9 layers · start here

The whole roadmap for understanding LLMs — from the math floor to the frontier — tying together every paper and course here into one guided path.

NeurIPS 2017

Attention Is All You Need

The Transformer: sequence transduction from attention alone — no recurrence, no convolutions. The architecture behind BERT, GPT, and every modern LLM.

28.4

BLEU EN→DE

8

attention heads

3.5 d

training time

ArchitectureNLPFoundational
SC 2024

Fire-Flyer AI-HPC & 3FS

DeepSeek's infrastructure playbook: 10,000 PCIe A100s, a CPU-assisted allreduce, and the 6.6 TiB/s 3FS file system — DGX-class training at half the cost.

≈50%

of DGX cost

6.6 TiB/s

3FS reads

10,000

A100 GPUs

SystemsHPCStorage
SIGMOD 2021

Real-time Data Infrastructure at Uber

How Uber turns trillions of Kafka messages a day into decisions in seconds — Kafka, Flink, Pinot, HDFS and Presto, re-engineered at petabyte scale. The blueprint for the modern real-time stack.

PBs

data per day

99.99%

availability SLA

< 1 s

p99 OLAP latency

SystemsStreamingOLAP

Primers

A whole field explained from zero — for when the papers assume background you do not have yet.

How Chips Get Made

6 sections · no prior knowledge

The semiconductor industry from zero: the chain from design to assembly, who sits where and who could actually be replaced, CPU vs GPU vs NPU vs ASIC, why one Dutch company is the chokepoint, and what export controls really block.

PrimerHardwareGeopolitics

Courses

Longer-form, interactive curricula built the same way as the papers.

LLM Deployment from First Principles

3 of 9 live

A nine-module course on serving large language models — from your laptop to hyperscale H100 clusters — for engineers who already know distributed systems. Live calculators, an autoregressive-loop animation, and an interactive roofline.

CourseInferenceSystems

Engineering handbooks

Runbooks for tools worth understanding properly — architecture first, then the commands, then the failure modes.

MiniStack + StackPort

8 sections · macOS + Ubuntu

Running ~80 AWS services on a single port locally, and browsing them: the six-signal request router, setup on macOS and Ubuntu, AWS CLI profiles that need no flags, where state actually lives — and a catalogue of every failure hit on a clean install.

Local AWSDockerRunbook

On the reading list

Candidates for the next interactive edition — links go to the originals for now.