caption

Author: zirkelc

README.md

@@ -19,21 +19,17 @@ The **llama** ([/ˈlɑːmə/](https://en.wikipedia.org/wiki/Help:IPA/English "He
 #### Markdown as Hierarchical Tree
 A properly structured markdown document forms a hierarchical tree where headings define sections containing various nodes (paragraphs, lists, tables, etc.). We parse markdown into an Abstract Syntax Tree (AST) and transform it into a hierarchical structure where sections contain their related content. This enables intelligent chunking that keeps semantically related information together.
 
-<figure>
 <img width="1266" height="542" alt="image" src="https://github.com/user-attachments/assets/0a49589c-fae2-4955-b042-5bee770f0344" />
-  <sup><sub>Hierarchical Markdown Abstract Syntax Tree</sub></sup>
-<figcaption>Hierarchical Markdown Abstract Syntax Tree</figcaption>
-</figure>
+
+  <small>Hierarchical Markdown Abstract Syntax Tree</small>
 
 
 #### Words as Atomic Unit
 Words are the smallest meaningful unit of information for embedding purposes. While tokenizers may split words further, for practical RAG applications, breaking words mid-way creates meaningless chunks. Therefore, words are treated as indivisible atoms that cannot be split.
 
-<figure>
-
 <img width="1263" height="282" alt="image" src="https://github.com/user-attachments/assets/65d00092-7bf4-4a15-94a9-7c80078e59eb" />
-<figcaption>Comparison of chunks with size 1 of Chunkdown (left) / LangChain Markdown Splitter (right)</figcaption>
-</figure>
+
+<small>Comparison of chunks with size 1 of Chunkdown (left) / LangChain Markdown Splitter (right)</small>
 
 #### Never Break Semantics
 Semantic elements like links, images, inline code, and certain formatting elements should ideally always remain intact. Breaking a long link like `[example](https://example.com)` into `[exam` and `ple](https://example.com)` destroys meaning. The splitter preserves these constructs even if it means exceeding the target chunk size.