---
license: cc-by-nc-4.0
task_categories:
- text-generation
- conversational
language:
- hi
- en
tags:
- hinglish- chat- companion- general-purpose- code-mixed
size_categories:
- 10K<n<100K
pretty_name: Hinglish Saathi Diverse
---
# Hinglish Saathi Diverse
## Dataset Description
A diverse Hinglish conversational dataset for a **General AI Companion ('Saathi')**. It contains a mix of deep multi-turn chats and quick single-turn replies, featuring code-mixed urban lingo and casual tones.
### Dataset Structure
- **Train:** 95% of data
- **Test:** 5% of data (held out for evaluation)
- **Format:** ShareGPT/OpenAI format (`messages` list with `user` and `assistant` roles).
### Language Style
- **Script:** Latin (Roman) script.
- **Grammar:** Hindi grammar with English technical terms.
- **Persona:** Companion focus.
### Source
- **Base Data:** Synthetic Generation (Diverse Personas)
- **Generator:** Gemma-3-27b-it (Quantized)
- **Method:** Structured Synthetic Generation via vLLM.
### License
**CC-BY-NC-4.0**
This dataset is derived from EmpatheticDialogues and synthetic prompts. It is intended for non-commercial research and educational use.