prakhar146/indian-finance-rag

Indian Finance RAG Dataset

Vector database for Indian financial RAG system.

Contents

File	Description
data/	HF Dataset with chunks + embeddings
faiss_index.bin	FAISS IndexFlatIP for fast retrieval
chunks.json	Raw text chunks with metadata
metadata.json	Dataset statistics

Stats

• Total chunks : 24,780
• Embedding dim: 384
• Embed model : all-MiniLM-L6-v2
• Chunk size : 400 chars with 80 overlap

Topics Covered

• NSE/BSE market structure
• RBI monetary policy
• SEBI regulations
• Mutual funds (direct/regular)
• Technical analysis
• Fundamental analysis
• Indian taxation (LTCG/STCG)
• Risk management
• F&O trading

Usage

from datasets import load_dataset
import faiss
import numpy as np
from huggingface_hub import hf_hub_download
from sentence_transformers import SentenceTransformer

# Load dataset
ds = load_dataset("prakhar146/indian-finance-rag", split="train")

# Load FAISS index
index_path = hf_hub_download(
    repo_id="prakhar146/indian-finance-rag",
    filename="faiss_index.bin",
    repo_type="dataset"
)
index = faiss.read_index(index_path)

# Search
model = SentenceTransformer("all-MiniLM-L6-v2")
query = "What is LTCG tax on equity?"
q_emb = model.encode([query], normalize_embeddings=True)
scores, indices = index.search(q_emb.astype(np.float32), k=3)
for i, idx in enumerate(indices[0]):
    print(f"Result {i+1}: {ds[int(idx)]['text'][:200]}")