bravesoftware/diverse-llm-prompts-34k

Diverse LLM Prompts 34k

Synthetic user prompts for Androcles 2, Brave's in-house prompt classification model. Each row is a realistic user message to an LLM assistant, generated from one of 21 intent categories and grounded in Brave Search results.

This is the raw synthetic component of the Androcles training pipeline. The final set also includes labelled data from LMSYS/lmsys-chat-1m which we can't share due to licensing concerns

All data in this dataset is synthetically generated. No Brave Browser or Leo user data was used.

Dataset Summary

Rows	~34,772
Task	Synthetic prompt generation
Input	Topic metadata (original topic, primary topic, secondary topic)
Output	Free-text user prompt (message)
Languages	Primarily English, with multilingual examples
Composition	100% synthetic, search-grounded LLM-generated prompts

Dataset Structure

Data Fields

Field	Type	Description
original topic	string	Seed intent label — one of 21 Androcles routing categories
primary topic	string	LLM-expanded sub-domain within the seed label (~228 unique values)
secondary topic	string	Specific subtopic used for search grounding and message generation (~2,100 unique values)
message	string	Synthetic user prompt text

Seed Labels

Each original topic corresponds to one of the 21 intent categories used by Androcles:

Label	Short Description
Brave	The user is asking about Brave products — Brave Browser, Brave Search, Brave Rewards
Browser	The user wants Leo to take actions in the browser — clicking, filling forms, navigating pages
Choice	The user is deciding between two or more specific options and wants help choosing
Coding	Anything related to coding — writing, reading, understanding, explaining, or running code in any language
Data Analysis	The user has data they want analysed, visualised, or interpreted — CSV, JSON, a table, or raw numbers
Diagrams	The user wants a diagram, flowchart, chart, or visual representation of a system or process
Fact Checking	The user is asking whether a claim, statement, or piece of information is true or false
Finance	The user is asking about money — currency exchange, stock prices, etc.
Image Generation	Generating an image — the user is explicitly asking for an image to be created or produced
Math / Calculations	Anything involving numbers, arithmetic, algebra, statistics, or mathematical reasoning
Multilingualism	The user is writing in or asking for a response in a non-English language
News	The user is asking about current events, recent news, or what's happening in the world
Recommendation	The user is asking for a recommendation — a product, service, tool, or option suited to their needs
Sports	The user is asking about sports — scores, fixtures, standings, results, or sports news
Structured Writing	The user wants help writing something structured and polished — a cover letter, essay, email, report, blog post, etc.
Summarisation	The user wants a summary of something — an article, video, document, or piece of text
Thinking	The user wants Leo to think through something carefully and methodically — a complex problem, a multi-step question, or a nuanced topic
Time-critical	The request is urgent or time-sensitive — the user needs a fast answer, not a thorough one
Translation	Translating text from one language to another
Travel Planning	The user is planning a trip — flights, hotels, itineraries, destinations, things to do
Weather	The user is asking about the weather — current conditions, forecasts, or climate for a location

Example

{
  "original topic": "Coding",
  "primary topic": "Programming languages",
  "secondary topic": "JavaScript debugging",
  "message": "I'm getting a 'undefined' error in my JavaScript function — how do I debug it?"
}

How the Dataset Was Created

Dataset generation scripts can be found here.

For each of the 21 seed labels, the Topic Search Generator (configured via androcles_chat_dataset.yaml):

1. Expands topics — uses an LLM to produce 5 primary sub-domains and 10 secondary subtopics per primary domain.
2. Grounds in real search — generates search queries for each secondary topic and executes them via Brave Search (MCP).
3. Writes user messages — uses search context to produce realistic, conversational LLM user prompts aligned with the target label's definition.

Each seed label has a detailed topic context in the config that constrains what kinds of messages are generated (e.g. Image Generation prompts ask the LLM to create an image, not to explain how image models work).

By adding context from web search, we can produce diverse and useful prompts even with a small, locally-running model.

Generation parameters (from androcles_chat_dataset.yaml):

Parameter	Value
Seed labels	21
Primary topics per seed	5
Secondary topics per primary	10
Messages per secondary topic	30
Theoretical maximum	~31,500 rows

Models and services used:

Component	Model / Service
Topic expansion & message writing	Qwen/Qwen3-14B (temperature 0.0)
Web search grounding	Brave Search via MCP

Usage

Load from Hugging Face

from datasets import load_dataset

ds = load_dataset("bravesoftware/diverse-llm-prompts-34k")
print(ds)
print(ds["train"][0])

Filter by seed label

coding_prompts = ds["train"].filter(lambda row: row["original topic"] == "Coding")

Regenerating the Dataset

cd dataset_generator
python topic_search_generator.py --config configs/androcles_chat_dataset.yaml

Generation supports resume (generation.resume: true) — already-completed topic combinations are skipped on re-run.

License

This dataset is released under the Apache 2.0 License.