Algorithms for AI

A structured, cross-linked DSA knowledge base that carries every topic through to where it shows up in AI systems: tries into tokenisation, heaps into beam search, graphs into GraphRAG, matrices into attention's O(n²) wall. Each guide has a paired practice page where Python runs in the browser.

Algorithm courses stop at the whiteboard, and LLM engineering material assumes the algorithms. The gap is the reason KV caches, HNSW indexes and beam search feel like magic to people who could implement a heap.

Connect the two: teach the classic structure, then show the exact place it lives inside a modern model or retrieval system, and let the reader run it.

Four sections — foundation, core DSA, advanced DSA and domain-specific DSA — with a runtime, a visualisation framework and a concept graph on top.

In-browser Python

Pyodide 0.26 loaded in a Web Worker with streamed stdout and stderr and on-demand package loading such as NumPy.

Visualisation framework

An immutable frame model for arrays, graphs, trees and matrices, renderers with play/pause/step controls and keyboard support, and frame builders for HNSW and KV-cache walkthroughs.

Concept graph

A knowledge graph of concepts with an explorer, learning paths and local progress tracking.

Domain-specific guides

ANN search, HNSW, tokenisation, matrix operations for attention, multi-head attention, KV cache, sampling and decoding, beam search, graph algorithms for GraphRAG, state machines and DAGs for LangGraph, KD and ball trees, streaming and caching, probabilistic structures, and AI systems benchmarks.

• Docusaurus 3.6 with TypeScript and React, Mermaid diagrams and client-side search.
• Content validators for links and concepts, eleven Vitest suites, and a deploy workflow gated on CI success.

1. 01

   Run it, don't just read it

   Every guide ships a practice page backed by a real Python runtime in the browser.

2. 02

   Deterministic frames

   Animations are built from immutable frames so each step is inspectable and testable.

Content

43 guides · 43 practice pages

Counted from the repository tree.

• Running a Python interpreter in a Web Worker without blocking the page or losing streamed output.

• At the time of writing the deployed site lags the repository: the live build serves the earlier 31-guide curriculum because the most recent CI runs failed before deployment.

• Interactive explanations need a runtime, not just illustrations.

• Repository (opens in a new tab)github.com/NiravRVaghasiya/dsa-knowledge
• Live site (opens in a new tab)niravrvaghasiya.github.io/dsa-knowledge/