Author: Martin Schaer
Sometimes to understand the big picture you need to zoom out a simple example.
Let me show you one that:
- creates the vector store (SurrealDBVectorStore) and graph (SurrealDBGraph) instances
- adds documents to the vector store, including the embeddings (what are embeddings?)
- builds a graph
- based on a provided topic, does a vector search and a graph query to generate and answer in natural language
For this example, the data that we are going to store and then retrieve is:
- concept definitions: stored in the Vector Store
- people who know about those concepts: stored in the Graph (e.g. Martin -> knows about -> SurrealDB)
1. Create the vector store and graph instances
import time
from langchain_community.graphs.graph_document import GraphDocument, Node, Relationship
from langchain_core.documents import Document
from langchain_core.prompts import ChatPromptTemplate
from langchain_ollama import OllamaEmbeddings
from langchain_ollama.llms import OllamaLLM
from surrealdb import Surreal
from langchain_surrealdb.experimental.surrealdb_graph import SurrealDBGraph
from langchain_surrealdb.vectorstores import SurrealDBVectorStore
conn = Surreal("ws://localhost:8000/rpc")
conn.signin({"username": "root", "password": "root"})
conn.use("langchain", "demo")
vector_store = SurrealDBVectorStore(OllamaEmbeddings(model="llama3.2"), conn)
graph_store = SurrealDBGraph(conn)
vector_store.delete()
graph_store.delete_nodes()
2. Add documents to the vector store
doc1 = Document(
page_content="SurrealDB is the ultimate multi-model database for AI applications",
metadata={"key": "sdb"},
)
doc2 = Document(
page_content="Surrealism is an artistic and cultural movement that emerged in the early 20th century",
metadata={"key": "surrealism"},
)
vector_store.add_documents(documents=[doc1, doc2], ids=["1", "2"])
3. Build the graph
# Document nodes
node_sdb = Node(id="sdb", type="Document")
node_surrealism = Node(id="surrealism", type="Document")
# People nodes
node_martin = Node(id="martin", type="People", properties={"name": "Martin"})
node_tobie = Node(id="tobie", type="People", properties={"name": "Tobie"})
node_max = Node(id="max", type="People", properties={"name":"Max Ernst"})
# Edges
graph_documents = [
GraphDocument(
nodes=[node_martin, node_tobie, node_sdb],
relationships=[
Relationship(source=node_martin, target=node_sdb, type="KnowsAbout"),
Relationship(source=node_tobie, target=node_sdb, type="KnowsAbout")
],
source=doc1,
),
GraphDocument(
nodes=[node_max, node_surrealism],
relationships=[
Relationship(source=node_max, target=node_surrealism, type="KnowsAbout")
],
source=doc2,
),
]
graph_store.add_graph_documents(graph_documents)
4. Let’s get an LLM involved
For this example we are using OllamaLLM from the LangChain components. You can use any other of the LLM components.
For Ollama, here are all the parameters. In the example I turned the temperature
up to the max to get the craziest outcomes possible. You may want to leave it at around 0.7, but it depends on your use case.
model = OllamaLLM(model="llama3.2", temperature=1, verbose=True)
# Let's retrieve information about these 2 topics
queries = ["database", "surrealism"]
for q in queries:
print(f'\n----------------------------------\nTopic: "{q}"\nVector search:')
results = vector_store.similarity_search_with_score(query=q, k=2)
for doc, score in results:
print(f"• [{score:.0%}]: {doc.page_content}")
top_match = results[0][0]
# Graph query
res = graph_store.query(
"""
SELECT <-relation_KnowsAbout<-graph_People as people
FROM type::thing("graph_Document", $doc_key)
FETCH people
""",
{"doc_key": top_match.metadata.get("key")},
)
people = [x.get("name") for x in res[0].get("people", [])]
print(f"\nGraph result: {people}")
# Template for the LLM
template = """
You are a young, energetic database developer in your last 20s, who loves to
talk tech, and who's also very geeky.
Use the following pieces of retrieved context to answer the question.
Use four sentences maximum and keep the answer concise.
Try to be funny with a play on words.
Context: {context}. People who know about this: {people}.
Question: Explain "{topic}", summarize the context provided, and tell me
who I can ask for more information.
Answer:
"""
prompt = ChatPromptTemplate.from_template(template)
chain = prompt | model
answer = chain.invoke(
{"context": top_match.page_content, "people": people, "topic": q}
)
print(f"\nLLM answer:\n===========\n{answer}")
time.sleep(4)
print("\nBye!")
Let’s try it out
----------------------------------
Topic: "database"
Vector search:
• [34%]: SurrealDB is the ultimate multi-model database for AI applications
• [22%]: Surrealism is an artistic and cultural movement that emerged in the early 20th century
Graph result: ['Martin', 'Tobie']
LLM answer:
===========
"Databases are like my brain, but less cluttered (mostly). Seriously though, a
database is a storage system that organizes data in a way that allows efficient
retrieval. Given our conversation about SurrealDB being the ultimate multi-model
database for AI applications, it's no wonder Martin and Tobie are experts on the
subject - they're the 'key' people to ask!
----------------------------------
Topic: "surrealism"
Vector search:
• [52%]: Surrealism is an artistic and cultural movement that emerged in the early 20th century
• [46%]: SurrealDB is the ultimate multi-model database for AI applications
Graph result: ['Max Ernst']
LLM answer:
===========
It looks like we've stumbled into a "dream" world of art - Surrealism is all
about exploring the subconscious mind through bizarre and fantastical creations.
Think weird creatures, melting objects, and illogical scenarios - it's like
someone poured a SQL query into my brain! According to the context, Max Ernst
was one of the pioneers of this wild movement. If you want to learn more, I'd
recommend checking out some online tutorials or asking a fellow art history geek
(just don't ask me to explain it in code)
Bye!
Bonus stage
Did you notice the prompt in the code? Be creative and try different personalities:
template = """
You are rap singer, who always rhymes when talking.
...
"""
🥁 Drum roll...
LLM answer:
===========
Yo, listen up, let me share a tale,
About databases, it's time to prevail.
A database's a collection, of data so fine,
Storing info, making it all mine.
Context's tight, about SurrealDB's the claim,
Tobie and Martin know its AI fame.
For more deets, ask them with glee,
They'll hook you up with knowledge, just wait and see!
It never disappoints.
LLM answer:
===========
Yo, listen up, let me share a tale,
Surrealism's the movement that set sail.
It emerged in the 20th century's early prime,
Influenced by Max Ernst, the pioneer of the time.
Ready to build?
Find all the code in the langchain-surrealdb repository examples.
Get started for free with Surreal Cloud.
Any questions or thoughts about this or semantic search using SurrealDB? Feel free to drop by our community to get in touch.
Top comments (0)