Chaitra

An open curriculum on scalable probabilistic systems — the sketches, protocols, and tradeoffs behind systems that answer “probably correct, much faster” instead of “exact, eventually.”

Learn online: chaitra.pages.dev · Browse the catalog: /catalog

---

What we are trying to do

At large scale, exact answers are often too slow, too large, or too coordinated. Production systems deliberately use probability: Bloom filters, HyperLogLog, gossip, power-of-two choices, approximate caches, rate limits, LLM token distributions, and more.

Chaitra is one repository that holds:

1. Deep narratives — each topic is a folder under topics/ with a README.md (math, diagrams, where it shows up in the wild).
2. Reference code — Python and/or Rust where we have implemented the idea (e.g. topics/BLOOM_FILTERS/bloom_filter.py).
3. A public website — static Astro app in website/ that syncs those READMEs, hosts interactive labs, and runs in-browser code (Pyodide, WASM where available).

We are building a guided path, not a wiki dump:

Layer	Where it lives	What you get
Story	topics/*/README.md → synced to the site	Intuition, formulas, Mermaid diagrams
Lab	website/src/components/ + topic MDX	Sliders, simulations, play/pause
Code	Repo sources + website Code workbench	Edit Python in the browser; read Rust; WASM demo on Bloom filters
Source of truth	This repo on GitHub	Clone, test, extend implementations

The long-form “textbook” content stays in topics/. This root README explains the project; each topics/TOPIC_NAME/README.md explains that topic.

---

How it works (repository model)

flowchart TB
  subgraph topicsDir [topics/]
    BF[BLOOM_FILTERS/]
    HLL[HYPERLOGLOG/]
    MORE[50+ other topics/]
  end
  subgraph each [Per topic]
    RM[README.md]
    CODE[.py / .rs optional]
  end
  subgraph site [website/]
    SYNC[sync:readme · sync:code · catalog]
    PAGES[chaitra.pages.dev]
  end
  BF --> RM
  BF --> CODE
  HLL --> RM
  MORE --> RM
  RM --> SYNC
  CODE --> SYNC
  SYNC --> PAGES

Loading

• One folder per topic — under topics/, SCREAMING_SNAKE_CASE (e.g. topics/COUNT_MIN_SKETCH/) with at least a README.md.
• Website slugs — kebab-case URLs (/topics/count-min-sketch) generated by website/scripts/generate-catalog.mjs.
• Catalog status — golden, live, preview, readme-only, coded (see website README).

Flagship starting point: Bloom filters — full story, lab, Python + Rust + WASM code workbench.

---

What are “scalable probabilistic systems”?

Traditional systems optimize for: same input → same output.

Probabilistic systems at scale optimize for: high-confidence answer, much lower cost — memory, coordination, latency, or compute.

That shows up as:

• Probabilistic data structures — Bloom filters, HyperLogLog, Count–Min Sketch
• Weaker coordination — gossip, eventual consistency, probabilistic load balancing
• Approximate policies — cache admission (TinyLFU), RED in networking, rate limiting
• ML / LLM stacks — token probabilities, MoE routing, large-scale inference

Each topic folder in topics/ goes deeper on one slice of that landscape. The site’s learning path orders a beginner-friendly subset; the catalog lists everything we track.

---

Repository layout

chaitra/
├── topics/                 # All curriculum modules (README + code)
│   ├── BLOOM_FILTERS/
│   ├── HYPERLOGLOG/
│   └── …
├── website/                # Astro site → Cloudflare Pages
│   ├── src/content/topics/ # Per-slug MDX + synced readme-body.md
│   └── README.md           # Build, deploy, contributor details
├── docs/
└── README.md               # You are here

---

For learners

1. Open chaitra.pages.dev and pick a topic from the catalog or learning path.
2. Read Story, try the Lab, experiment in Code where available.
3. Clone this repo and open the matching folder (e.g. topics/BLOOM_FILTERS/) to run Python or Rust locally.

git clone https://github.com/07AMIT10/chaitra.git
cd chaitra/topics/BLOOM_FILTERS
python bloom_filter.py

---

For contributors

Add or improve a topic

1. Edit or create topics/YOUR_TOPIC/README.md (and add .py / .rs if you have reference code).

2. If the topic should appear on the site, add or update website/src/content/topics/<slug>/ (index.mdx, optional code.mdx, lab component).

3. From website/:

   npm run sync:readme    # after README changes
   npm run sync:code      # after Rust snippets used in Code tabs
   npm run catalog        # refresh topics.json
   npm run build

4. Open a PR — Cloudflare can preview website/ changes.

Site development, deploy, CSP, WASM: see website/README.md and website/DEPLOY-CLOUDFLARE.md.

CI: pushes touching website/** or topics/** run the site build via .github/workflows/website-ci.yml.

---

Topic index (on GitHub)

Every module with a README.md lives under topics/. Highlights with labs and/or code on the site:

Topic folder	Site
topics/BLOOM_FILTERS/	/topics/bloom-filters
topics/HYPERLOGLOG/	/topics/hyperloglog
topics/COUNT_MIN_SKETCH/	/topics/count-min-sketch
topics/CONSISTENT_HASHING/	/topics/consistent-hashing
topics/GOSSIP_PROTOCOLS/	/topics/gossip-protocols

For the full list and status badges, use the catalog (generated from topics/).

---

Status

Early but active: many topics have rich READMEs; labs and code workbench coverage grow topic by topic. Issues and PRs welcome.

---

Links

• Live site: https://chaitra.pages.dev
• Curriculum index: topics/README.md
• Website docs: website/README.md
• Deploy: website/DEPLOY-CLOUDFLARE.md