adm_ai_rag_chunk_api

Chunking API for RAG implementations Provides reusable text chunking functionality with smart break points and support for markdown headers

Default chunking parameters:

• Target chunk size: 2048 characters (approximately 512 tokens)
• Minimum chunk size: 512 characters (25% of target)
• Maximum chunk size: 2560 characters (150% of target)
• Overlap size: 256 characters (12.5% of target)

Functions and Procedures

find_break_point

Find the next good break point in content Prefers: paragraph breaks, code blocks, sentence endings, then newlines

Signature:

function find_break_point (
  p_content in clob,
  p_start_pos in number,
  p_target_pos in number,
  p_max_pos in number
) return number;

Parameters:

Name	Direction	Type	Description
p_content	in	clob	The content to search in
p_start_pos	in	number	The starting position for the search
p_target_pos	in	number	The target position for the break
p_max_pos	in	number	The maximum position allowed

Returns: number - Position where the break should occur

---

find_header_positions

Find header positions in markdown content Returns table of positions where headers start

Signature:

function find_header_positions (
  p_content in clob
) return sys.odcinumberlist;

Parameters:

Name	Direction	Type	Description
p_content	in	clob	The markdown content to analyze

Returns: sys.odcinumberlist - Collection of header positions

---

find_html_heading_positions

Find HTML heading positions (h1-h6 tags) Returns table of positions where headings start

Signature:

function find_html_heading_positions (
  p_content in clob
) return sys.odcinumberlist;

Parameters:

Name	Direction	Type	Description
p_content	in	clob	The HTML content to analyze

Returns: sys.odcinumberlist - Collection of heading positions

---

find_html_tag_positions

Find HTML tag boundaries (major structural tags like div, section, article, table) Returns table of positions where these tags start

Signature:

function find_html_tag_positions (
  p_content in clob
) return sys.odcinumberlist;

Parameters:

Name	Direction	Type	Description
p_content	in	clob	The HTML content to analyze

Returns: sys.odcinumberlist - Collection of tag positions

---

find_json_structure_positions

Find JSON object/array boundaries at root level Returns table of positions where top-level structures start

Signature:

function find_json_structure_positions (
  p_content in clob
) return sys.odcinumberlist;

Parameters:

Name	Direction	Type	Description
p_content	in	clob	The JSON content to analyze

Returns: sys.odcinumberlist - Collection of boundary positions

---

find_xml_element_positions

Find XML element boundaries (major elements) Returns table of positions where major elements start

Signature:

function find_xml_element_positions (
  p_content in clob
) return sys.odcinumberlist;

Parameters:

Name	Direction	Type	Description
p_content	in	clob	The XML content to analyze

Returns: sys.odcinumberlist - Collection of element positions

---

find_paragraph_positions

Find paragraph boundaries in plain text content Returns table of positions where paragraphs start

Signature:

function find_paragraph_positions (
  p_content in clob
) return sys.odcinumberlist;

Parameters:

Name	Direction	Type	Description
p_content	in	clob	The text content to analyze

Returns: sys.odcinumberlist - Collection of paragraph positions

---

get_chunking_strategy

Determine chunking strategy based on MIME type Returns: ‘markdown_headers’, ‘html_tags’, ‘json_structure’, ‘xml_elements’, ‘paragraphs’, or ‘default’

Signature:

function get_chunking_strategy (
  p_mime_type in varchar2
) return varchar2;

Parameters:

Name	Direction	Type	Description
p_mime_type	in	varchar2	The MIME type of the content

Returns: varchar2 - The chunking strategy to use

---

create_rag_chunks

Create RAG chunks for content with MIME type-aware chunking strategy Automatically selects appropriate chunking strategy based on content type:

• Markdown: respects headers (h1-h6)
• HTML: respects major structural tags (div, section, article, table)
• JSON: respects root-level objects/arrays
• XML: respects major element boundaries
• Plain text: respects paragraph boundaries
• Other: uses character-based smart breaking

Signature:

procedure create_rag_chunks (
  p_file_id in adm_ai_rag_collection_files.rag_collection_file_id%type,
  p_target_chunk_size in number default 2048,
  p_min_chunk_size in number default 512,
  p_max_chunk_size in number default 2560,
  p_overlap_size in number default 256
);

Parameters:

Name	Direction	Type	Description
p_file_id	in	adm_ai_rag_collection_files.rag_collection_file_id%type	The RAG collection file ID
p_target_chunk_size	in	number default 2048	Target chunk size in characters (default: 2048)
p_min_chunk_size	in	number default 512	Minimum chunk size in characters (default: 512)
p_max_chunk_size	in	number default 2560	Maximum chunk size in characters (default: 2560)
p_overlap_size	in	number default 256	Overlap size in characters for adjacent chunks (default: 256)

---