baguswicak/tani-bot-transcripts

TaniBot YouTube Transcripts Dataset

Indonesian Agriculture YouTube Video Transcripts for RAG Pipeline

---

📊 Dataset Info

• Videos: 3 (sample)
• Transcript Chunks: 10
• Languages: Indonesian (Bahasa), English
• Topics: Pest control, organic farming, planting techniques, crop rotation
• Total Duration: ~30 minutes of content

---

🎯 Purpose

This dataset contains transcribed content from Indonesian agriculture YouTube videos, chunked and formatted for RAG (Retrieval-Augmented Generation) pipelines.

Use Cases:

• RAG retrieval for TaniBot chatbot
• Training data for agricultural Q&A systems
• Knowledge base for farmer assistance
• Multi-lingual agricultural information access

---

📁 Schema

Field	Type	Description
id	string	Unique chunk identifier (e.g., trans_0000_chunk_001)
video_id	string	YouTube video ID
video_title	string	Video title
channel_name	string	YouTube channel name
chunk_id	integer	Chunk number within video
chunk_text	string	Transcript text (~1000 characters)
language	string	Language code (id or en)
duration_seconds	integer	Video duration in seconds
topics	list[string]	Topic tags
source_url	string	Full YouTube URL

---

🚀 Usage

Load with Hugging Face Datasets

from datasets import load_dataset

# Stream (recommended for large datasets)
dataset = load_dataset("baguswicak/tani-bot-transcripts", split="train", streaming=True)

# Or load fully
dataset = load_dataset("baguswicak/tani-bot-transcripts", split="train")

# Iterate
for chunk in dataset:
    print(f"Video: {chunk['video_title']}")
    print(f"Text: {chunk['chunk_text'][:200]}...")

Filter by Language

# Indonesian only
id_chunks = dataset.filter(lambda x: x['language'] == 'id')

# English only
en_chunks = dataset.filter(lambda x: x['language'] == 'en')

Filter by Topic

# Pest control chunks
pest_chunks = dataset.filter(lambda x: 'pest' in x['topics'])

---

📺 Video Sources

1. Cara Mengatasi Hama pada Tanaman Padi dan Jagung

• Channel: Petani Sukses
• Language: Indonesian
• Topics: Pest control, rice, corn, IPM
• Chunks: 5

2. Amazing Rice Farming Technique from Indonesia

• Channel: Farming Innovation
• Language: English
• Topics: Rice planting, Jajar Legowo technique
• Chunks: 2

3. Tutorial Pertanian Organik: Pestisida Nabati

• Channel: Kebun Organik
• Language: Indonesian
• Topics: Organic farming, natural pesticide, DIY
• Chunks: 3

---

🔄 Updating the Dataset

To add more transcripts:

1. Scrape new videos:

   python scripts/youtube_transcript_scraper.py
   

2. Add video IDs to script:

   VIDEO_IDS = {
       "new_video_id": {
           "title": "...",
           "channel": "...",
           "topics": ["..."],
           "language": "id"
       }
   }
   

3. Upload to Hugging Face:

   from huggingface_hub import HfApi
   api = HfApi()
   dataset.push_to_hub("baguswicak/tani-bot-transcripts")
   

---

🔗 Integration with TaniBot RAG

from supabase import create_client
from datasets import load_dataset

# Load transcripts
transcripts = load_dataset("baguswicak/tani-bot-transcripts", split="train")

# Combine with Supabase RAG
supabase = create_client(SUPABASE_URL, SUPABASE_KEY)

# Search Supabase embeddings
similar_docs = supabase.rpc('match_documents', {
    'query_embedding': query_embedding,
    'threshold': 0.7,
    'count': 5
}).execute()

# Combine contexts
context = similar_docs.data + [t['chunk_text'] for t in transcripts.take(3)]

---

📈 Scaling Plan

Current: 10 chunks (3 videos, ~30 min)
Week 2 Target: 50 chunks (10 videos, ~2 hours)
Month 2 Target: 500 chunks (100 videos, ~20 hours)

Topic Expansion:

• ✅ Pest & disease management
• ✅ Organic farming
• ✅ Planting techniques
• ⏳ Fertilizer guides
• ⏳ Harvest & post-harvest
• ⏳ Weather adaptation
• ⏳ Business & marketing

---

📝 Chunking Strategy

Method: Sentence-based splitting
Max Chunk Size: 1000 characters
Overlap: None (clean breaks at sentence boundaries)

Why this works:

• Each chunk is self-contained
• Fits within LLM context windows
• Easy to retrieve and combine
• Preserves semantic meaning

---

⚠️ Notes

• Transcripts are auto-generated by YouTube (may contain errors)
• Some videos may not have transcripts available
• Indonesian language videos may have mixed Bahasa/English
• Chunks are not manually verified (future work)

---

📄 License

CC BY-SA 4.0

Attribution:

• Original video content: Respective YouTube creators
• Transcript processing: TaniBot Project

---

🔗 Links

• GitHub: https://github.com/wizzleweasel/tani-bot
• Hugging Face: https://huggingface.co/datasets/baguswicak/tani-bot-transcripts
• TaniBot: https://huggingface.co/spaces/baguswicak/tani-bot

---

Last Updated: 2026-04-28
Maintained By: TaniBot Team