Implementation:Explodinggradients Ragas PromptUtils Module

Field	Value
source	Explodinggradients_Ragas (GitHub)
domains	Prompts, Utilities
last_updated	2026-02-10 00:00 GMT

Overview

The prompt.utils module provides utility functions for recursively extracting strings from nested structures, replacing strings in those structures, and extracting JSON from LLM text output.

Description

get_all_strings recursively traverses any combination of strings, Pydantic models, lists, tuples, and dictionaries to collect all string values into a flat list. update_strings performs the inverse operation: given parallel lists of old and new strings, it deep-copies the structure and replaces each exact string match with its corresponding new string. Both functions are essential for the prompt adaptation/translation workflow. extract_json identifies the first JSON structure (object or array) in a text blob by matching balanced delimiters. It handles markdown-wrapped JSON (triple-backtick blocks) by starting search after the ```json marker. If no balanced JSON structure is found, the original text is returned.

Usage

Use get_all_strings and update_strings together for prompt translation workflows. Use extract_json to parse LLM outputs that contain JSON embedded in natural language text.

Code Reference

Item	Detail
Source Location	src/ragas/prompt/utils.py L7-106
Functions	get_all_strings(obj) -> list[str], update_strings(obj, old_strings, new_strings) -> Any, extract_json(text) -> str
Import	from ragas.prompt.utils import get_all_strings, update_strings, extract_json

I/O Contract

Inputs

Function	Parameter	Type	Description
get_all_strings	obj	Any	Nested structure (str, BaseModel, list, tuple, dict)
update_strings	obj	Any	Nested structure to update
update_strings	old_strings	list[str]	Strings to find (parallel with new_strings)
update_strings	new_strings	list[str]	Replacement strings (parallel with old_strings)
extract_json	text	str	Raw LLM output text potentially containing JSON

Outputs

Function	Return Type	Description
get_all_strings	list[str]	Flat list of all string values found
update_strings	Any	Deep copy of obj with strings replaced
extract_json	str	Extracted JSON string, or original text if no JSON found

Usage Examples

from ragas.prompt.utils import get_all_strings, update_strings, extract_json

# Extract all strings from nested structures
data = {"a": "hello", "b": ["world", "foo"], "c": {"d": "bar"}}
strings = get_all_strings(data)
print(strings)  # ['hello', 'world', 'foo', 'bar']

# Replace strings in nested structures
updated = update_strings(
    data,
    old_strings=["hello", "world", "foo", "bar"],
    new_strings=["hola", "mundo", "baz", "qux"],
)
print(updated)  # {'a': 'hola', 'b': ['mundo', 'baz'], 'c': {'d': 'qux'}}

# Extract JSON from LLM output
text = 'Here is the result: {"score": 0.85, "reason": "good"} hope that helps!'
json_str = extract_json(text)
print(json_str)  # '{"score": 0.85, "reason": "good"}'

# Handles markdown code blocks
md_text = '```json\n{"answer": "yes"}\n```'
print(extract_json(md_text))  # '{"answer": "yes"}'

Related Pages

• PydanticPrompt_Class - Uses extract_json for output parsing and get_all_strings/update_strings for adapt()
• MetricBasePrompt_Class - Uses get_all_strings/update_strings for adapt()
• BasePrompt_Class - Base prompt classes that rely on these utilities