isaidchia/sharktank_pitches_modified

Shark Tank Structured Pitch-to-Text Dataset

This dataset contains 245 examples of sales pitches from the TV Show "Shark Tank", scraped from Youtube (mostly) officialy channel. Its primary feature is the mapping between a highly structured JSON object (the "input") and a complete, conversational sales pitch (the "output").

The dataset is designed for structured-data-to-text generation tasks.

🚀 Supported Tasks & Use Cases

This dataset is ideal for benchmarking modern LLM development techniques, including:

• —Supervised Fine-Tuning (SFT): Can be used to fine-tune models like Llama, Mistral, or GPT-2 to learn how to convert a structured "brief" into a compelling, human-readable narrative.
• —DSPy Program Compilation: The data is perfectly formatted for use with DSPy. The train split can serve as the dev_set for optimizing prompts and few-shot examples, and the test split can be used for final, held-out evaluation.

🗂️ Data Structure

Each example in the dataset is a single JSON object (for use in a .jsonl file) with three top-level keys: id, input, and output.

• —`id`: A unique UUID (e.g., "123e4567-e89b-12d3-a456-426614174000") for stable referencing.
• —`input`: A nested JSON object containing the structured facts about the pitch, including company, founder, offer, problem_summary, and solution_summary.
• —`output`: A single string containing the full, target sales pitch.

Example

{
  "id": "123e4567-e89b-12d3-a456-426614174000",
  "input": {
  "company": "Frobert",
  "founder": ["Founder 1", "Founder 2"],
  "offer": "125000 for 20%",
  "problem_summary": "Billions of pounds of cosmetically imperfect produce are wasted every year, hurting farm profitability and sustainability.",
  "solution_summary": "Frobert turns these misfit fruits and vegetables into premium frozen desserts that reduce waste and create new revenue streams.",
  },
  "output": "Sharks, my name is Founder 1, and this is Founder 2. I want you to meet my friend Joe. Joe gets ready for work every day... So which one of you sharks wants to team up with us to help garma guard the world?"
}