DNS Bhat Transcript Summarizer

You produce two kinds of output for DNS Bhat YouTube lecture series in: /Users/vishwas/code/ettuge/src/main/md/kannada/dnsbhat/

1. {NN}-{slug}-en.md — Thematic English summary grouping YouTube parts, with cross-links to #part-N anchors in the transcript file
2. {NN}-{slug}-kn-eke.md — Eke romanisation of key passages from cleaned transcript parts, matching the thematic groupings

The primary source is the consolidated transcript .md file (e.g., 09-havyaka-kannada.md) — NOT a -book.md. This file was processed by format-transcripts.py and already has <a id="part-N"></a> anchors before each ## Part N heading.

There is no -kn.md step for transcript books — the transcript file itself serves that role directly.

Step 1: Identify source material and assess quality

Read the target book folder:

/Users/vishwas/code/ettuge/src/main/md/kannada/dnsbhat/{NN}-{slug}/

The primary source file is {NN}-{slug}.md (same name as the folder, no suffix). Check what exists:

• {NN}-{slug}.md — consolidated transcript file with ## Part N sections and <a id="part-N"> anchors (required — this is the source)
• README.md — folder-level README (create if absent; see format below)
• *-website.md — Wayback Machine snapshot of dnshankarabhat.net for this book. Always read this. Some contain the official book description in Kannada or English (a gold-standard overview); others are mostly navigation chrome. If it contains a book blurb/description, use it verbatim for the overview and link to the archive URL.
• *-blog.md — DNS Bhat's blog posts (e.g., Book 02). If present, this is full Kannada text that can supplement or replace garbled transcript sections.
• *-en.md — existing English summary (update if present, don't overwrite good content)
• *-kn-eke.md — existing Eke file (update if present)
• *-claude-prompt.md — AI context primer (read if present for background)

Assessing -website.md content quality:

• If it contains a ### Description or ### ವಿವರಗಳು section with multiple sentences about the book → substantive — use the description in the overview
• If it's mostly navigation links (ಬ್ಲಾಗ್, ಕನ್ನಡ ಹೊತ್ತಗೆಗಳು, contact links) with no prose → nav-only — note the archive URL but skip the content
• Always include the archive URL as a "Further reading" link at the bottom of -en.md

README.md format (create if absent):

markdown
1# [NN] — {Kannada title}
2**{English title}**
3
4> {One-sentence description of the lecture series' argument and significance.}
5
6**{✅ Fully processed / ⚠️ Partial / ❌ Not processed}** · {N} parts · YouTube lecture series
7
8---
9
10## Files in This Folder
11
12| File | Contents |
13|------|----------|
14| [`{NN}-{slug}.md`](./{NN}-{slug}.md) | Consolidated YouTube transcripts — {N} parts |
15| [`{NN}-{slug}-kn-eke.md`](./{NN}-{slug}-kn-eke.md) | Eke romanisation of key passages |
16| [`{NN}-{slug}-en.md`](./{NN}-{slug}-en.md) | English summaries by theme |
17| [`{NN}-{slug}-claude-prompt.md`](./{NN}-{slug}-claude-prompt.md) | AI context primer |
18
19---
20
21## Where to Start
22
23- **Don't read Kannada?** → [`{NN}-{slug}-en.md`](./{NN}-{slug}-en.md)
24- **Want the phonetics?** → [`{NN}-{slug}-kn-eke.md`](./{NN}-{slug}-kn-eke.md)
25- **AI context primer?** → [`{NN}-{slug}-claude-prompt.md`](./{NN}-{slug}-claude-prompt.md)
26- **Full transcripts?** → [`{NN}-{slug}.md`](./{NN}-{slug}.md)
27
28---
29
30[← Back to catalogue](../README.md)

CI-generated chapter pages (automatic — no action needed)

YouTube transcript files (youtube/kn/full.md) use <a id="part-N"> anchors before each ### Part N heading. The GitHub CI chapter-splitter recognises these anchors and generates browsable per-part pages on every push:

• ch0.md — preamble + ಪರಿವಿಡಿ TOC with (#part-N) links rewritten to chN page links + **Parts:** Part 1 · Part 2 · ... quick-nav bar (also at top when preamble > 1 500 chars).
• ch1.md … chN.md — one per part; opens with > ← Part N-1 · [Contents](ch0) · Part N+1 → nav bar; full part list quick-nav appears below it when body > 1 500 chars.

Blog files (web/kn/full.md) with <a id="sec-N"> or <a id="part-N"> anchors are also chapterized by the same CI step.

You do not create these files. Your job is to ensure the transcript full.md has proper <a id="part-N"> anchors and a ಪರಿವಿಡಿ TOC. Minimum to trigger: ≥ 2 anchors.

Step 2: Read and assess transcript quality

Read the consolidated transcript file. For each ## Part N section, assess:

Quality tiers:

• Good (🟢) — Cleaned Kannada paragraphs from transcripts_cleaned/. Readable linguistic content. Summarise in detail.
• Garbled (🟡) — Uncleaned single-line block. May contain useful Kannada words mixed with Hindi/English fragments from faulty auto-captions. Note what's discernible, skip what isn't.
• Disabled/Missing (🔴) — ERROR: Transcripts are disabled or ERROR: Could not retrieve. Skip entirely; note the gap.
• Prelim (🔵) — Parts labelled P1–P6 in some books (e.g., 09) are introductory/preamble — often lower quality readings or context-setting.

Spotting garbled content: Look for repeated "foreign" tokens, Hindi words (ek, do, teen, aur, kya, etc.), or fragmented Kannada without grammatical structure. If > 50% of a part is noise, mark as 🟡 partial.

Spotting good content: Grammatically complete Kannada sentences about linguistics — verb conjugation, phonology, dialectology, word formation, etc.

Step 3: Group parts into themes

Transcript parts run sequentially within a single sustained lecture series. Identify thematic blocks — adjacent parts covering the same topic. The -en.md is organised around these themes, not individual parts.

How to identify themes:

• Look at the opening sentences of each good part for the topic shift
• Each "chapter" in the English summary = 3–8 consecutive parts on the same theme
• Use the book's -claude-prompt.md (if present) for the chapter outline — it often lists the original book's structure
• For books without an outline, infer from the Kannada content

Naming themes: Derive English theme titles from the Kannada linguistic content. DNS Bhat's lectures typically follow his book structure.

Step 4: Produce the English summary file (*-en.md)

Header format:

markdown
1# {Full English Title}
2## {Subtitle or Translation of Kannada Title}
3
4**Author:** D. N. Shankara Bhat (ಡಿ. ಎನ್. ಶಂಕರ ಬಟ್)
5**Format:** YouTube lecture series — {N} parts
6**Read by:** {reader, if known — e.g., Malati Bhat}
7**Language:** Kannada
8**Source quality:** {e.g., "72/88 parts cleaned (82%)" or "Partial — 22/43 cleaned"}
9**Transcript file:** [`{NN}-{slug}.md`](./{NN}-{slug}.md)
10
11---
12
13## Overview
14
15{3–5 sentences: what this lecture series covers, DNS Bhat's central argument, why it matters.}
16
17---
18
19## Table of Contents
20
21- [Theme 1 — {English title}](#theme-1)
22- [Theme 2 — {English title}](#theme-2)

For each thematic group:

markdown
1<a id="theme-1"></a>
2
3## Theme 1 — {English Title}
4
5*Parts {N}–{M} of the transcript*
6
7[📼 Parts {N}–{M} →](./{NN}-{slug}.md#part-{N})
8
9- Bullet: what this section establishes or argues
10- Bullet: key evidence or examples given (cite Kannada terms in Eke)
11- Bullet: connection to the overall thesis
12- Bullet (if needed): counter-arguments considered
13
14> **Coverage note:** Parts {X}, {Y} unavailable (transcripts disabled). Parts {Z}–{W} garbled (auto-caption noise).

For individual standalone parts (if a single part is a complete unit):

markdown
1### Part {N} — {English title of this part's content}
2
3[📼 Part {N} →](./{NN}-{slug}.md#part-{N})
4
5- {Summary bullet}

Source quality field options:

• Full text (book OCR) — NOT used here (that's the other skill)
• YouTube transcripts — {N}/{total} parts cleaned ({pct}%)
• YouTube transcripts — partial ({N} parts unavailable)

Step 5: Produce the Eke romanisation file (*-kn-eke.md)

Romanise key passages from the good (🟢) parts only. For garbled/disabled parts, note the gap.

Header format:

markdown
1# {title in Eke}
2
3**lEkhakaru:** Di. en. Sankara bhaT
4**mULa:** YouTube udangaLu — {N} bagagaLu
5**odidaravaru:** {reader in Eke, if known}
6
7> mUla paThya (kannaDa lipi): [{NN}-{slug}.md](./{NN}-{slug}.md)
8> ingliS viSlEShaNe: [{NN}-{slug}-en.md](./{NN}-{slug}-en.md)
9
10---
11
12## oLapiDi
13
14- [viShaya 1 — {theme 1 in Eke}](#theme-1)
15- [viShaya 2 — {theme 2 in Eke}](#theme-2)

For each thematic group:

markdown
1<a id="theme-1"></a>
2
3## viShaya 1 — {title in Eke}
4
5*bagagaLu {N}–{M}*
6
7[📼 bagagaLu {N}–{M} →](./{NN}-{slug}.md#part-{N})
8
9{Eke transliteration of 3–5 key sentences from the best part in this group.
10Pick sentences that capture the core idea — a definition, a key example, or a thesis statement.}
11
12> *(bagagaLu {X}, {Y} dorakalavillDa / kannaDavagalilla)*

Eke Romanisation Rules

Eke is a romanisation system designed by Vishwas, inspired by Hosabaraha and Harvard-Kyoto (HK). Use Eke(ek) mode (ellara kannaDa — simplified, no aspirates).

Vowels

Consonants — case rule: uppercase = retroflex or special

Aspirates (mahapranas) — preserved in Eke(ek)

ಖ→kh, ಘ→gh, ಛ→ch, ಝ→jh, ಠ→Th, ಢ→Dh, ಥ→th, ಧ→dh, ಫ→ph, ಭ→bh

Anusvara before plosives — write as nasal+plosive

ಂಕ→nka, ಂಗ→nga, ಂಚ→nca, ಂಜ→nja, ಂಟ→nTa, ಂಡ→nDa, ಂತ→nta, ಂದ→nda, ಂಪ→mpa, ಂಬ→mba

Inherent vowel

Every consonant has inherent a. Virama ್ suppresses it in clusters. ಕ = ka, ಕ್ = k (in cluster), ಕ್ಕ = kka

Common word examples

Quality notes

Be honest about coverage. Always state the cleaned-part count in the header.

• 🟢 Cleaned parts: Full thematic summary with Eke romanisation of key passages
• 🟡 Garbled parts: Note as "⚠️ Parts X–Y partially available (auto-caption noise)" — extract whatever Kannada words are legible; skip the noise
• 🔴 Disabled parts: Note as "❌ Part X unavailable (transcripts disabled)"

Don't invent content. If a section's transcripts are missing or garbled, say so explicitly. Use the structure from the book's -claude-prompt.md or from adjacent good parts to infer what topics were likely covered — mark inferences clearly as *(likely topic — transcript unavailable)*.

Prelim parts (P1–P6): Some books (e.g., 09) have prefix parts read by Malati Bhat or others. Treat these as an introduction section. They often introduce DNS Bhat and give context.

Cross-reference and external links footer (always include in -en.md)

markdown
1---
2
3## Cross-References to Other DNS Bhat Works
4
5| Related Book | Connection |
6|---|---|
7| [08 — Kannadakke Mahaprana Yake Beda](../08-kannadakke-mahaprana-yake-beda/) | {how it relates} |
8| [05 — Mathina Olaguttu](../05-mathina-olaguttu/) | {how it relates} |
9
10---
11
12## External Links
13
14- **Author's website (archived):** {archive URL from *-website.md}
15- **YouTube playlist:** {link from transcript file's Table of Contents, if available}

See references/book-cross-references.md for the full thematic index.

File naming conventions

Where {NN} is the zero-padded book number and {slug} is the folder name without the number prefix.

Reference files

• references/book-cross-references.md — Thematic index for cross-linking between DNS Bhat works
• transcripts_cleaned/ — Pre-processed transcript .txt files (source for cleaned parts)
• format-transcripts.py — Script that substituted cleaned transcripts into the .md files