Update sample code for the article on LangChain chatbot

langchain-rag-app/.gitignore

@@ -0,0 +1,4 @@
+__pycache__/
+*.pyc
+.env
+chroma_data/

langchain-rag-app/README.md

@@ -4,41 +4,56 @@ This repo contains the source code for [Build an LLM RAG Chatbot With LangChain]
 
 To run the final application that you'll build in this tutorial, you can use the code provided in `source_code_final/`.
 
+## Project Layout
+
+Each folder holds the code as it stands at the end of the matching step in the tutorial:
+
+| Folder | Tutorial step |
+| --- | --- |
+| `data/` | The hospital system CSV files used throughout the tutorial |
+| `source_code_step_1/` | Step 1: Understand the Business Requirements and Data |
+| `source_code_step_2/` | Step 2: Set Up a Neo4j Graph Database |
+| `source_code_step_3/` | Step 3: Build a Graph RAG Chatbot in LangChain |
+| `source_code_step_4/` | Step 4: Deploy the LangChain Agent |
+| `source_code_final/` | The finished application |
+
+The LangChain basics that this tutorial builds on live in a separate tutorial, [LangChain Tutorial: Build Your First Chains and Agents](https://realpython.com/langchain-tutorial/), along with its own sample code.
+
 ## Setup
 
 Create a `.env` file in the root directory and add the following environment variables:
 
 ```.env
-OPENAI_API_KEY=<YOUR_OPENAI_API_KEY>
-
 NEO4J_URI=<YOUR_NEO4J_URI>
 NEO4J_USERNAME=<YOUR_NEO4J_USERNAME>
 NEO4J_PASSWORD=<YOUR_NEO4J_PASSWORD>
 
-HOSPITALS_CSV_PATH=https://raw.githubusercontent.com/hfhoffman1144/langchain_neo4j_rag_app/main/data/hospitals.csv
-PAYERS_CSV_PATH=https://raw.githubusercontent.com/hfhoffman1144/langchain_neo4j_rag_app/main/data/payers.csv
-PHYSICIANS_CSV_PATH=https://raw.githubusercontent.com/hfhoffman1144/langchain_neo4j_rag_app/main/data/physicians.csv
-PATIENTS_CSV_PATH=https://raw.githubusercontent.com/hfhoffman1144/langchain_neo4j_rag_app/main/data/patients.csv
-VISITS_CSV_PATH=https://raw.githubusercontent.com/hfhoffman1144/langchain_neo4j_rag_app/main/data/visits.csv
-REVIEWS_CSV_PATH=https://raw.githubusercontent.com/hfhoffman1144/langchain_neo4j_rag_app/main/data/reviews.csv
+HOSPITALS_CSV_PATH=https://raw.githubusercontent.com/realpython/materials/refs/heads/master/langchain-rag-app/data/hospitals.csv
+PAYERS_CSV_PATH=https://raw.githubusercontent.com/realpython/materials/refs/heads/master/langchain-rag-app/data/payers.csv
+PHYSICIANS_CSV_PATH=https://raw.githubusercontent.com/realpython/materials/refs/heads/master/langchain-rag-app/data/physicians.csv
+PATIENTS_CSV_PATH=https://raw.githubusercontent.com/realpython/materials/refs/heads/master/langchain-rag-app/data/patients.csv
+VISITS_CSV_PATH=https://raw.githubusercontent.com/realpython/materials/refs/heads/master/langchain-rag-app/data/visits.csv
+REVIEWS_CSV_PATH=https://raw.githubusercontent.com/realpython/materials/refs/heads/master/langchain-rag-app/data/reviews.csv
 
-HOSPITAL_AGENT_MODEL=gpt-3.5-turbo-1106
-HOSPITAL_CYPHER_MODEL=gpt-3.5-turbo-1106
-HOSPITAL_QA_MODEL=gpt-3.5-turbo-0125
+OPENAI_API_KEY=<YOUR_OPENAI_API_KEY>
+HOSPITAL_AGENT_MODEL=gpt-5.6-luna
+HOSPITAL_CYPHER_MODEL=gpt-5.6-luna
+HOSPITAL_QA_MODEL=gpt-5.6-luna
+HOSPITAL_EMBEDDING_MODEL=text-embedding-3-small
 
-CHATBOT_URL=http://host.docker.internal:8000/hospital-rag-agent
+CHATBOT_URL=http://chatbot_api:8000/hospital-rag-agent
 ```
 
 The chatbot uses OpenAI LLMs, so you'll need to create an [OpenAI API key](https://realpython.com/generate-images-with-dalle-openai-api/#get-your-openai-api-key) and store it as `OPENAI_API_KEY`. 
 
-The three `NEO4J_` variables are used to connect to your Neo4j AuraDB instance. Follow the directions [here](https://neo4j.com/cloud/platform/aura-graph-database/?ref=docs-nav-get-started) to create a free instance.
+The three `NEO4J_` variables configure the Neo4j Community Edition instance that Docker Compose runs for you, so there's no cloud account to set up. Point `NEO4J_URI` at the Bolt endpoint on your machine, use the default `neo4j` user, and choose your own password of at least eight characters. Compose reads `NEO4J_PASSWORD` when it creates the database container, and the ETL and chatbot services reach Neo4j over the Compose network instead of `localhost`.
 
-Once you have a running Neo4j instance, and have filled out all the environment variables in `.env`, you can run the entire project with [Docker Compose](https://docs.docker.com/compose/). You can install Docker Compose by following [these directions](https://docs.docker.com/compose/install/).
+Keep your real credentials in `.env` only. That file is listed in `.gitignore`, so it stays out of version control.
 
-Once you've filled in all of the environment variables, set up a Neo4j AuraDB instance, and installed Docker Compose, open a terminal and run:
+Once you've filled in all of the environment variables and installed [Docker Compose](https://docs.docker.com/compose/install/), open a terminal and run:
 
 ```console
-$ docker-compose up --build
+$ docker compose up --build
 ```
 
 After each container finishes building, you'll be able to access the chatbot API at `http://localhost:8000/docs` and the Streamlit app at `http://localhost:8501/`.

langchain-rag-app/source_code_final/chatbot_api/Dockerfile

@@ -1,12 +1,10 @@
-# chatbot_api/Dockerfile
-
-FROM python:3.11-slim
+FROM python:3.14-slim
 
 WORKDIR /app
 COPY ./src/ /app
 
 COPY ./pyproject.toml /code/pyproject.toml
-RUN pip install /code/.
+RUN python -m pip install /code/.
 
 EXPOSE 8000
-CMD ["sh", "entrypoint.sh"]
+CMD ["sh", "entrypoint.sh"]

langchain-rag-app/source_code_final/chatbot_api/pyproject.toml

@@ -2,18 +2,16 @@
 name = "chatbot_api"
 version = "0.1"
 dependencies = [
-    "asyncio==3.4.3",
-    "fastapi==0.109.0",
-    "langchain==0.1.0",
-    "langchain-openai==0.0.2",
-    "langchainhub==0.1.14",
-    "neo4j==5.14.1",
-    "numpy==1.26.2",
-    "openai==1.7.2",
-    "opentelemetry-api==1.22.0",
-    "pydantic==2.5.1",
-    "uvicorn==0.25.0"
+    "fastapi==0.136.3",
+    "langchain==1.3.4",
+    "langchain-openai==1.2.2",
+    "langchain-neo4j==0.9.0",
+    "neo4j==6.2.0",
+    "numpy==2.4.6",
+    "openai==2.41.0",
+    "pydantic==2.13.4",
+    "uvicorn==0.49.0"
 ]
 
 [project.optional-dependencies]
-dev = ["black", "flake8"]
+dev = ["ruff"]

langchain-rag-app/source_code_final/chatbot_api/src/agents/hospital_rag_agent.py

@@ -1,23 +1,46 @@
 import os
 
+from langchain.agents import create_agent
+from langchain_core.tools import Tool
+from langchain_openai import ChatOpenAI
+
 from chains.hospital_cypher_chain import hospital_cypher_chain
 from chains.hospital_review_chain import reviews_vector_chain
-from langchain import hub
-from langchain.agents import AgentExecutor, Tool, create_openai_functions_agent
-from langchain_openai import ChatOpenAI
 from tools.wait_times import (
     get_current_wait_times,
     get_most_available_hospital,
 )
 
 HOSPITAL_AGENT_MODEL = os.getenv("HOSPITAL_AGENT_MODEL")
 
-hospital_agent_prompt = hub.pull("hwchase17/openai-functions-agent")
+agent_system_prompt = (
+    "You are a helpful assistant for a hospital system. Use the tools "
+    "available to you to answer the user's questions about patients, "
+    "visits, physicians, hospitals, insurance payers, patient reviews, "
+    "and current wait times."
+)
+
+
+def query_reviews(query: str) -> str:
+    """Answer questions about patient experiences from their reviews."""
+    try:
+        return reviews_vector_chain.invoke(query)
+    except Exception as e:
+        return f"The patient reviews are unavailable right now: {e}"
+
+
+def query_graph(query: str) -> str:
+    """Answer questions by querying the hospital graph database."""
+    try:
+        return hospital_cypher_chain.invoke(query)["result"]
+    except Exception as e:
+        return f"The hospital database couldn't answer that: {e}"
+
 
 tools = [
     Tool(
         name="Experiences",
-        func=reviews_vector_chain.invoke,
+        func=query_reviews,
         description="""Useful when you need to answer questions
         about patient experiences, feelings, or any other qualitative
         question that could be answered about a patient using semantic
@@ -30,7 +53,7 @@
     ),
     Tool(
         name="Graph",
-        func=hospital_cypher_chain.invoke,
+        func=query_graph,
         description="""Useful for answering questions about patients,
         physicians, hospitals, insurance payers, patient review
         statistics, and hospital visit details. Use the entire prompt as
@@ -63,20 +86,10 @@
     ),
 ]
 
-chat_model = ChatOpenAI(
-    model=HOSPITAL_AGENT_MODEL,
-    temperature=0,
-)
-
-hospital_rag_agent = create_openai_functions_agent(
-    llm=chat_model,
-    prompt=hospital_agent_prompt,
-    tools=tools,
-)
+chat_model = ChatOpenAI(model=HOSPITAL_AGENT_MODEL, use_responses_api=True)
 
-hospital_rag_agent_executor = AgentExecutor(
-    agent=hospital_rag_agent,
+hospital_rag_agent_executor = create_agent(
+    model=chat_model,
     tools=tools,
-    return_intermediate_steps=True,
-    verbose=True,
+    system_prompt=agent_system_prompt,
 )

langchain-rag-app/source_code_final/chatbot_api/src/chains/hospital_cypher_chain.py

@@ -1,8 +1,7 @@
 import os
 
-from langchain.chains import GraphCypherQAChain
-from langchain.prompts import PromptTemplate
-from langchain_community.graphs import Neo4jGraph
+from langchain_core.prompts import PromptTemplate
+from langchain_neo4j import GraphCypherQAChain, Neo4jGraph
 from langchain_openai import ChatOpenAI
 
 HOSPITAL_QA_MODEL = os.getenv("HOSPITAL_QA_MODEL")
@@ -50,7 +49,8 @@
 # Which physician has billed the least to Cigna
 MATCH (p:Payer)<-[c:COVERED_BY]-(v:Visit)-[t:TREATS]-(phy:Physician)
 WHERE p.name = 'Cigna'
-RETURN phy.name AS physician_name, SUM(c.billing_amount) AS total_billed
+RETURN phy.id AS physician_id, phy.name AS physician_name,
+       SUM(c.billing_amount) AS total_billed
 ORDER BY total_billed
 LIMIT 1
 
@@ -64,33 +64,37 @@
      v.admission_date < '2023-01-01' THEN 1 ELSE 0 END) AS count_2022,
      SUM(CASE WHEN v.admission_date >= '2023-01-01' AND
      v.admission_date < '2024-01-01' THEN 1 ELSE 0 END) AS count_2023
+WHERE count_2022 > 0
 WITH state, visit_count, count_2022, count_2023,
      (toFloat(count_2023) - toFloat(count_2022)) / toFloat(count_2022) * 100
      AS percent_increase
 RETURN state, percent_increase
 ORDER BY percent_increase DESC
 LIMIT 1
 
-# How many non-emergency patients in North Carolina have written reviews?
-match (r:Review)<-[:WRITES]-(v:Visit)-[:AT]->(h:Hospital)
-where h.state_name = 'NC' and v.admission_type <> 'Emergency'
-return count(*)
+# How many reviews are there for non-emergency visits at North Carolina
+# hospitals?
+MATCH (r:Review)<-[:WRITES]-(v:Visit)-[:AT]->(h:Hospital)
+WHERE h.state_name = 'NC' and v.admission_type <> 'Emergency'
+RETURN count(*)
 
 String category values:
 Test results are one of: 'Inconclusive', 'Normal', 'Abnormal'
 Visit statuses are one of: 'OPEN', 'DISCHARGED'
 Admission Types are one of: 'Elective', 'Emergency', 'Urgent'
-Payer names are one of: 'Cigna', 'Blue Cross', 'UnitedHealthcare', 'Medicare',
+Payer names are one of: 'Cigna', 'Blue Cross', 'UnitedHealthcare', 'Medicaid',
 'Aetna'
 
 A visit is considered open if its status is 'OPEN' and the discharge date is
 missing.
 Use abbreviations when
 filtering on hospital states (e.g. "Texas" is "TX",
 "Colorado" is "CO", "North Carolina" is "NC",
-"Florida" is "FL", "Georgia" is "GA, etc.)
+"Florida" is "FL", "Georgia" is "GA", etc.)
 
 Make sure to use IS NULL or IS NOT NULL when analyzing missing properties.
+Use duration.inDays(date1, date2).days when counting the number of days
+between two dates.
 Never return embedding properties in your queries. You must never include the
 statement "GROUP BY" in your query. Make sure to alias all statements that
 follow as with statement (e.g. WITH v as visit, c.billing_amount as
@@ -109,7 +113,7 @@
 qa_generation_template = """You are an assistant that takes the results
 from a Neo4j Cypher query and forms a human-readable response. The
 query results section contains the results of a Cypher query that was
-generated based on a users natural language question. The provided
+generated based on a user's natural language question. The provided
 information is authoritative, you must never doubt it or try to use
 your internal knowledge to correct it. Make the answer sound like a
 response to the question.
@@ -128,15 +132,14 @@
 results are in units of days unless otherwise specified.
 
 When names are provided in the query results, such as hospital names,
-beware  of any names that have commas or other punctuation in them.
+beware of any names that have commas or other punctuation in them.
 For instance, 'Jones, Brown and Murray' is a single hospital name,
 not multiple hospitals. Make sure you return any list of names in
 a way that isn't ambiguous and allows someone to tell what the full
 names are.
 
 Never say you don't have the right information if there is data in
-the query results. Make sure to show all the relevant query results
-if you're asked.
+the query results. Always use the data in the query results.
 
 Helpful Answer:
 """
@@ -146,12 +149,13 @@
 )
 
 hospital_cypher_chain = GraphCypherQAChain.from_llm(
-    cypher_llm=ChatOpenAI(model=HOSPITAL_CYPHER_MODEL, temperature=0),
-    qa_llm=ChatOpenAI(model=HOSPITAL_QA_MODEL, temperature=0),
+    cypher_llm=ChatOpenAI(model=HOSPITAL_CYPHER_MODEL),
+    qa_llm=ChatOpenAI(model=HOSPITAL_QA_MODEL),
     graph=graph,
     verbose=True,
     qa_prompt=qa_generation_prompt,
     cypher_prompt=cypher_generation_prompt,
     validate_cypher=True,
     top_k=100,
+    allow_dangerous_requests=True,
 )

langchain-rag-app/source_code_final/chatbot_api/src/chains/hospital_review_chain.py

@@ -1,19 +1,26 @@
+import logging
 import os
 
-from langchain.chains import RetrievalQA
-from langchain.prompts import (
+from langchain_core.output_parsers import StrOutputParser
+from langchain_core.prompts import (
     ChatPromptTemplate,
     HumanMessagePromptTemplate,
     PromptTemplate,
     SystemMessagePromptTemplate,
 )
-from langchain.vectorstores.neo4j_vector import Neo4jVector
-from langchain_openai import ChatOpenAI, OpenAIEmbeddings
+from langchain_core.runnables import RunnablePassthrough
+from langchain_neo4j import Neo4jVector
+from langchain_openai import ChatOpenAI
+from langchain_openai import OpenAIEmbeddings
+
+# Silence Neo4j's deprecation notice for db.index.vector.queryNodes
+logging.getLogger("neo4j.notifications").setLevel(logging.ERROR)
 
 HOSPITAL_QA_MODEL = os.getenv("HOSPITAL_QA_MODEL")
+HOSPITAL_EMBEDDING_MODEL = os.getenv("HOSPITAL_EMBEDDING_MODEL")
 
 neo4j_vector_index = Neo4jVector.from_existing_graph(
-    embedding=OpenAIEmbeddings(),
+    embedding=OpenAIEmbeddings(model=HOSPITAL_EMBEDDING_MODEL),
     url=os.getenv("NEO4J_URI"),
     username=os.getenv("NEO4J_USERNAME"),
     password=os.getenv("NEO4J_PASSWORD"),
@@ -29,11 +36,10 @@
 )
 
 review_template = """Your job is to use patient
-reviews to answer questions about their experience at
-a hospital. Use the following context to answer questions.
-Be as detailed as possible, but don't make up any information
-that's not from the context. If you don't know an answer,
-say you don't know.
+reviews to answer questions about their experience at a hospital. Use
+the following context to answer questions. Be as detailed as possible,
+but don't make up any information that's not from the context. If you
+don't know an answer, say you don't know.
 {context}
 """
 
@@ -52,9 +58,12 @@
     input_variables=["context", "question"], messages=messages
 )
 
-reviews_vector_chain = RetrievalQA.from_chain_type(
-    llm=ChatOpenAI(model=HOSPITAL_QA_MODEL, temperature=0),
-    chain_type="stuff",
-    retriever=neo4j_vector_index.as_retriever(k=12),
+reviews_retriever = neo4j_vector_index.as_retriever(search_kwargs={"k": 12})
+review_chat_model = ChatOpenAI(model=HOSPITAL_QA_MODEL)
+
+reviews_vector_chain = (
+    {"context": reviews_retriever, "question": RunnablePassthrough()}
+    | review_prompt
+    | review_chat_model
+    | StrOutputParser()
 )
-reviews_vector_chain.combine_documents_chain.llm_chain.prompt = review_prompt

langchain-rag-app/source_code_final/chatbot_api/src/entrypoint.sh

@@ -4,4 +4,4 @@
 echo "Starting hospital RAG FastAPI service..."
 
 # Start the main application
-uvicorn main:app --host 0.0.0.0 --port 8000
+uvicorn main:app --host 0.0.0.0 --port 8000