Using a Pareto approach with an LLM to learn Devanagari, Part 2.

Good morning, Gemini.

Let us refresh in our respective minds what we did with the approach to the Devanagari script last time, so that we might continue where we left off, in the same vein.

Good morning! We are officially back in the polyglot strategy room.

Let’s refresh our mental workspace and look at the foundation we built for our huliganov.tv masterclass.

Where We Left Off: The One-Hour Masterclass

We completely abandoned the traditional, overwhelming method of memorizing the alphabet from to . Instead, we deployed a high-frequency spy-code approach, targeting the letters most commonly used to write international names and loanwords in modern Hindi.

Here is what we unlocked in our first hour:

1. The High-Frequency Core (10 Consonants)

We mastered roughly 30% of the consonant lineup, which statistically unlocked over 80% of the sounds needed for global maps and transliterations:

  • The Liquids/Nasals: (), (), (), ()

  • The Hard Stops (Tongue Curled Back): (Hard ), (Hard )

  • The Global Heavyweights: (), (), (), ()

2. The Vowel Dashboard (The Matras)

We learned how Devanagari behaves as an abugida—where every standalone consonant has a built-in “ah” sound (schwa) that drops naturally at the very end of a word. We learned how to override that built-in vowel using graphic modifiers:

  • (Long ), ि (Short – written to the left), (Long ), / (/), / (Pure /).

  • Our Modern Upgrades: The English short vowel hacks—the Crescent Moon (ॉ) for the “O” in Boston/Coffee and the Double Feather (ै) for the “A” in Bank/Texas.

3. The Grand Finale: The 94% Literacy Proof

We put our toolkit to the test against all 50 US States. The result? A 94% functional literacy rate. Our readers could perfectly read or highly accurately guess almost the entire map (Texas = टेक्सस, Boston = बॉस्टन, California = कैलिफ़ोर्निया) using just that one hour of concepts.

Moving Forward: Writing Mechanics & Disguised Letters

Now, as we discussed, if we want to move our readers from a 67% overall capability toward that 100/100 total fluency point, we need to tackle how the letters are actually built.

 

In Devanagari, knowing the equivalent of stroke order isn’t just about good handwriting—it is a vital tool for optical recognition. Many letters look incredibly similar at a glance, and understanding which stroke is drawn first helps the brain instantly tell them apart.

The Golden Rule of Writing Devanagari

Unlike Western script where you write along a baseline, or Chinese characters drawn inside an imaginary box, Devanagari is built on a hanging line system.

  1. You write the distinctive curves and body of the letter first.

  2. You draw the vertical “anchor wall” (if the letter has one).

  3. Crucial Step: You draw the horizontal top bar (the shirorekha) last, effectively hanging the completed word from the line.

Are you ready to select the first cluster of letters from our high-frequency squad to analyze their stroke paths and compare them to the lookalikes that might confuse our readers in the next pass?

Continue reading “Using a Pareto approach with an LLM to learn Devanagari, Part 2.”

Using a Pareto approach with an LLM to learn Devanagari, Part 1

Using a Pareto approach with an LLM to learn Devanagari!

This article is basically an experiment in how linguists or polyglots or language enthusiasts (whatever we want to call ourselves, what does language matter 😉 ) can use LLMs, in this case Gemini on a PRO sub – you can try your own favorite one with the same prompts and see if it comes up with the same sort of thing) to learn something which is perhaps not altogether straightforward, but nevertheless very useful. In this case the writing system used, after Latin, Chinese characters and Arabic-based writing, by most people on the planet. The premise here is not to learn either Hindi or Sanskrit or any of the other languages that use Devanagari as languages. This is an exercise purely in getting to grips with a writing system. For this reason, we are looking at how this writing system transcribes international words, such as geographical or personal names, but not looking at any Hindi or Nepali per se, not every phrasebook phrases. Apart from anything else, the article should be as agnostic as possible with regard to which language form the set is of most interst to readers, whilst being aware that statistically Hindi will account for the greater part.

The prompts (that is, my bits) are in the H4 format, which for those unfamiliar with doing WordPress means the fourth header level format, which is what I am writing in now. What Gemini answers is in the other formats.

So hopefully you find this interesting on two levels: firstly the experiment as to whether, right now, in July 2026, I can simply dispense with an expensive book and ask Gemini to write me an instruction book on a topic like this as we go along, allowing me to dictate the order of play, the style and other matters, and the second point of interest is the content itself, that is, the actual learning of what in this article is actually only 20 of what it takes to know Devanagari inside out, but it’s been guided to follow a Pareto approach. The test at the close proves that Pareto has not entirely held true, because we have a 67:20 relation rather than a 80:20 as Wilf Pareto might have predicted based on his principle, but anyway I think it’s a good deal to give my readers 67% of a very useful skill while asking them to invest only one hour to gain it.

Obviously it will take a few iterations to get it well memorised, but each run at it is likely to be doable within an hour.

Here goes.

Continue reading “Using a Pareto approach with an LLM to learn Devanagari, Part 1”