Upload README.md with huggingface_hub

README.md

@@ -37,14 +37,30 @@ The full 1.8 GB dataset is available as a **Gated Repository** for commercial an
 
 ---
 
-## Sample Details
-- **Records:** 5,000 diverse records.
-- **Languages:** Trilingual (Sinhala, Tamil, English) + Singlish & Tanglish.
-- **Format:** OpenAI-style `messages` JSON Lines (`.jsonl`).
-
-## Intended Use
-- **Format Verification:** Test the tool-calling and reasoning format with your training pipeline.
-- **Data Quality Audit:** Verify the linguistic nuance and logical depth of the Chat2Find data.
+## 📊 Sample Details & Composition
+
+The 5,000 records in this preview are carefully curated to reflect the high quality of the full dataset. 
+
+**Conversation Flow:**
+- **Single-turn (SFT):** ~70.0% (3,500 records)
+- **Multi-turn (Agentic/Chat):** ~30.0% (1,500 records)
+
+**Reasoning & Execution:**
+- **Pure Chain-of-Thought Reasoning:** ~72.0%
+- **Tool Calling & API Interaction:** ~28.0%
+
+**Language Breakdown:**
+- **Tamil:** ~45.7%
+- **Sinhala:** ~36.3%
+- **English:** ~18.0%
+- *Note: Singlish and Tanglish code-mixed data are aggressively embedded within these records to ensure realistic South Asian conversational abilities.*
+
+---
+
+## 🌟 What's in the Full Dataset?
+The full 1.8 GB dataset contains **279,260 records** offering a massive scale-up of everything seen in this sample.
+- Over **10,800** deep multi-turn interactions.
+- Hundreds of thousands of localized, culturally aware logic puzzles, tool invocations, and code-mixed conversations not found in standard open-source datasets.
 
 ---
 **Stay tuned to [chat2find.com](https://chat2find.com) for updates.**