A year of AI-coding discourse taught me one thing: the hardest part isn't the tools — it's the vocabulary wall. If English isn't your first language, "token", "context window", "hallucination", "harness" don't sound like concepts. They sound like spells. 🪄
Matt Pocock's brilliant Dictionary of AI Coding showed how teachable these words really are. So I adapted it into Bangla — and then kept going, way past where the original stops. Today it's a complete, free, open-source learning path:
🔗 https://github.com/dhrupo/dictionary-of-ai-coding-bangla
The four-stage arc: শব্দ → হাতিয়ার → গল্প → অভ্যাস
(words → tools → story → habit)
📖 Part 1 — The words. 62 terms across 7 sections, each with a daily-life analogy instead of a textbook definition. A model is a calculator that never presses its own buttons. Parameters are mixing-board knobs. And there's গোল্ডি (Goldie) — a goldfish with a 3-second memory who keeps reminding you that models are stateless too. 🐠
🛠️ Part 2 — The tools. This is where it goes beyond the original: 7 hands-on guides covering real CLI commands (Claude Code + Codex, mapped back to the concepts — /compact is just the Compaction button), AI-friendly folder structure (AGENTS.md, .claude/, who reads what and when), bad-prompt → good-prompt rewrites, the popular community skills (brainstorming, grill-me, TDD, handoff…), 17 extra terms (RAG, embeddings, temperature, hooks…), token economics, and safety (including prompt injection, explained with a postman analogy).
📜 Part 3 — The story. Instead of a tutorial you have to install things for, it's a 6-chapter read-along: রাফি, a 9th grader, builds his first portfolio site with an AI agent. He makes every classic mistake — vague prompts, trusting a hallucinated library, dragging a session deep into the dumb zone — and recovers using the tools from Part 2. Margin notes call back every concept at the exact moment it bites. The reader installs nothing.
🗓️ The 30-day path. Reading isn't owning. So the book ends with 30 daily missions (5–15 min each). The first two weeks need zero setup — just any free AI chat and open eyes ("ask the same question twice — did the answers match?"). Weeks 3–4 go hands-on, but every mission has a read-only alternative.
Plus: a myth-busting FAQ ("AI will eat my job", "you need a CS degree"), a 79-term alphabetical index, copy-paste templates/ (a starter AGENTS.md, a working /handoff command, a SKILL.md example), and 26 terminal GIFs — all reproducible from committed VHS scripts.
Things I learned building it
- Analogies are load-bearing, not decoration. "Prefix cache" never landed until it became "the waiter remembers your usual order."
-
A story teaches sequencing in a way lists can't. You can document
/compactvs/clearvs handoff perfectly and people still won't know when. Watching রাফি lose his color-scheme decision to a long session — that sticks. -
Bengali text breaks your tooling in fun ways.
grepcouldn't match half my content (decomposed Unicode), GitHub slugs for emoji-prefixed headings start with a stray hyphen, and nested code fences need 4-backtick outers. The repo's link checker is a tiny Python script for a reason. - Beginners deserve the real thing. Simplified ≠ dumbed down. The dictionary keeps the exact English terms visible everywhere, so readers can walk into any English doc afterward and feel at home.
It's a community thing
Everything is CC BY 4.0. There are Bangla issue forms for "I want a new word" and "I found a mistake" — contributions of any size welcome, from a typo to a better analogy to a whole new section.
And if you speak another language: steal this structure. The vocabulary wall exists in every non-English-speaking dev community, and the fix is surprisingly fun to build. 💙
Top comments (0)