RAG Impl no internet access, locally running model

README.md

RAG implementation example

This is a RAG implementation example with no internet based API or tools.

Single-file RAG pipeline using:

• LlamaIndex (open source)
• Local Ollama (llama3 for both embeddings + generation)
• SingleStore DB (local vector store)

No internet, no paid APIs required.

How to run

1. Make sure Ollama is running with llama3: ollama pull llama3 && ollama run llama3
2. Have SingleStore running locally (Docker example): docker run -d -p 3306:3306 --name singlestore singlestore/cluster-in-a-box (create the database rag_db manually or let the script handle it)
3. Install dependencies (once):Bash
pip install llama-index llama-index-embeddings-ollama llama-index-llms-ollama llama-index-vector-stores-singlestore singlestoredb
4. Update the CONFIGURATION section with your paths/credentials.
5. python rag_pipeline.py

⠀The script now runs the complete pipeline end-to-end in one file. Enjoy!

no-internet-RAG-impl.py

#!/usr/bin/env python3
"""
Single-file RAG pipeline using:
- LlamaIndex (open source)
- Local Ollama (llama3 for both embeddings + generation)
- SingleStore DB (local vector store)
No internet, no paid APIs required.
"""

import os
from llama_index.core import (
    Settings,
    SimpleDirectoryReader,
    VectorStoreIndex,
    StorageContext,
)
from llama_index.core.node_parsers import SentenceSplitter
from llama_index.embeddings.ollama import OllamaEmbedding
from llama_index.vector_stores.singlestore import SingleStoreVectorStore
from llama_index.llms.ollama import Ollama

# ========================= CONFIGURATION =========================
# Change these to match your environment
DOCS_DIR = "/path/to/your/documents"          # ← your PDF/TXT/DOCX folder
OLLAMA_BASE_URL = "http://localhost:11434"    # default Ollama port
SINGLESTORE_CONFIG = {
    "host": "localhost",
    "port": 3306,
    "user": "root",
    "password": "your_password",              # ← change this
    "database": "rag_db",                     # ← change this
    "table": "vector_table",
}
EMBEDDING_MODEL = "llama3"                    # must be pulled with `ollama pull llama3`
LLM_MODEL = "llama3"
CHUNK_SIZE = 1024
CHUNK_OVERLAP = 20
TOP_K = 3

# Sample query (change as needed)
QUERY = "What is the main topic discussed in these documents?"

# ========================= STEP 1: LOAD & SPLIT =========================
print("🔄 Step 1: Loading and splitting documents...")
documents = SimpleDirectoryReader(
    input_dir=DOCS_DIR,
    required_exts=[".pdf", ".txt", ".docx", ".doc"],  # supported formats
).load_data()

parser = SentenceSplitter(chunk_size=CHUNK_SIZE, chunk_overlap=CHUNK_OVERLAP)
nodes = parser.get_nodes_from_documents(documents)

print(f"✅ Loaded {len(documents)} documents → {len(nodes)} nodes")

# ========================= STEP 2: EMBED & INDEX IN SINGLESTORE =========================
print("🔄 Step 2: Setting up embeddings + SingleStore vector store...")

Settings.embed_model = OllamaEmbedding(
    model_name=EMBEDDING_MODEL,
    base_url=OLLAMA_BASE_URL,
)

vector_store = SingleStoreVectorStore(
    host=SINGLESTORE_CONFIG["host"],
    port=SINGLESTORE_CONFIG["port"],
    user=SINGLESTORE_CONFIG["user"],
    password=SINGLESTORE_CONFIG["password"],
    database=SINGLESTORE_CONFIG["database"],
    table=SINGLESTORE_CONFIG["table"],
    embedding_dimension=4096,  # llama3 dimension
)

storage_context = StorageContext.from_defaults(vector_store=vector_store)

index = VectorStoreIndex(nodes, storage_context=storage_context)

print("✅ Documents embedded and indexed in SingleStore")

# ========================= STEP 3: RETRIEVAL =========================
print("\n🔄 Step 3: Retrieval demo...")
retriever = index.as_retriever(similarity_top_k=TOP_K)
retrieved_nodes = retriever.retrieve(QUERY)

print(f"Top {TOP_K} retrieved nodes:")
for i, node in enumerate(retrieved_nodes, 1):
    print(f"  {i}. {node.text[:200]}...")

# ========================= STEP 4: AUGMENT + GENERATE =========================
print("\n🔄 Step 4: Augment & Generate with local Ollama...")
Settings.llm = Ollama(
    model=LLM_MODEL,
    base_url=OLLAMA_BASE_URL,
    temperature=0.7,
    request_timeout=120.0,
)

query_engine = index.as_query_engine(similarity_top_k=TOP_K)

response = query_engine.query(QUERY)

print("\n" + "="*60)
print("FINAL RAG RESPONSE")
print("="*60)
print(response)
print("="*60)

# Optional: show source nodes used for the final answer
print("\nSources used:")
for i, node in enumerate(response.source_nodes, 1):
    print(f"  {i}. {node.node.metadata.get('file_name', 'unknown')} "
          f"(score: {node.score:.3f})")