jenngang's picture
Upload app.py with huggingface_hub
6fdd408 verified
# --- 0. Library Imports ---
import os
import json
import shutil
import time
import numpy as np
from datetime import datetime
from typing import Dict, List, Any, TypedDict, Tuple
# LangChain Core & Community
from langchain_core.documents import Document
from langchain_core.runnables import RunnablePassthrough
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.tools import tool
from langchain_community.vectorstores import Chroma
from langchain_community.document_loaders import PyPDFDirectoryLoader
from langchain_community.cross_encoders import HuggingFaceCrossEncoder
from langchain.text_splitter import CharacterTextSplitter, RecursiveCharacterTextSplitter
# LangChain OpenAI
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
# LangChain Experimental
from langchain_experimental.text_splitter import SemanticChunker
# LangChain Agents & Graph
from langchain.agents import create_tool_calling_agent, AgentExecutor
from langgraph.graph import StateGraph, END, START
# External Libraries
import chromadb
from llama_parse import LlamaParse # For PDF parsing
from groq import Groq # For Llama Guard
from mem0 import MemoryClient # For memory
import streamlit as st # For Web UI
# Fix for numpy depreciation warning if necessary
np.float_ = np.float64
#import nest_asyncio
# --- 1. Configuration and Setup Utilities ---
from typing import Dict, List, Any, TypedDict, Tuple
def load_config_from_env() -> Dict:
"""Loads API keys and endpoints from environment variables."""
# Prioritize environment variables for deployment
config = {
"AZURE_OPENAI_API_KEY": os.getenv("AZURE_OPENAI_API_KEY"),
"AZURE_OPENAI_API_BASE": os.getenv("AZURE_OPENAI_API_BASE"),
"LLAMA_KEY": os.getenv("LLAMA_KEY"), # LlamaParse API Key
"MEM0_API_KEY": os.getenv("MEM0_API_KEY"),
"GROQ_API_KEY": os.getenv("GROQ_API_KEY"), # For Llama Guard via Groq
}
# Basic validation
for key, value in config.items():
if not value:
st.warning(f"Warning: Environment variable '{key}' is not set.")
return config
def initialize_llms_and_embeddings(config: Dict) -> Tuple[OpenAIEmbeddings, ChatOpenAI, chromadb.utils.embedding_functions.OpenAIEmbeddingFunction, Groq]:
"""Initializes LLM, Embedding models, and API clients."""
api_key = config["AZURE_OPENAI_API_KEY"]
endpoint = config["AZURE_OPENAI_API_BASE"]
groq_api_key = config["GROQ_API_KEY"]
# Initialize ChromaDB embedding function (used for collection creation)
embedding_function = chromadb.utils.embedding_functions.OpenAIEmbeddingFunction(
api_base=endpoint,
api_key=api_key,
model_name='text-embedding-ada-002' # Specify model explicitly
)
# Initialize LangChain OpenAI Embeddings (used for `SemanticChunker` and `Chroma` vectorstore)
embedding_model = OpenAIEmbeddings(
openai_api_base=endpoint,
openai_api_key=api_key,
model='text-embedding-ada-002' # Specify model explicitly
)
# Initialize LangChain Chat OpenAI model
llm = ChatOpenAI(
openai_api_base=endpoint,
openai_api_key=api_key,
model="gpt-4o-mini",
streaming=False,
temperature=0.0 # Ensure deterministic behavior for evaluations
)
# Initialize Groq client for Llama Guard
llama_guard_client = Groq(api_key=groq_api_key)
return embedding_model, llm, embedding_function, llama_guard_client
def filter_input_with_llama_guard(user_input: str, llama_guard_client: Groq, model: str = "meta-llama/llama-guard-4-12b") -> str:
"""
Filters user input using Llama Guard to ensure it is safe.
Returns "safe", "UNSAFE" (with category), or None on error.
"""
try:
response = llama_guard_client.chat.completions.create(
messages=[{"role": "user", "content": user_input}],
model=model,
)
return response.choices[0].message.content.strip()
except Exception as e:
st.error(f"Error with Llama Guard: {e}")
return None
# --- 2. Data Preparation (Parsing & Chunking) ---
# Note: In a deployed app, PDF parsing and vector DB creation would typically be
# a separate, offline process, and the pre-built vector DB would be loaded.
# For this template, we'll assume the nutritional_db is pre-existing and loaded.
def load_and_split_documents(folder_path: str, embedding_model: OpenAIEmbeddings) -> List[Document]:
"""Loads PDF documents from a folder and semantically chunks them."""
semantic_text_splitter = SemanticChunker(
embedding_model,
breakpoint_threshold_type='percentile',
breakpoint_threshold_amount=80
)
pdf_loader = PyPDFDirectoryLoader(folder_path)
chunks = pdf_loader.load_and_split(semantic_text_splitter)
return chunks
def parse_pdf_tables_with_llamaparse(pdf_path: str, llamaparse_api_key: str) -> Tuple[Dict, Dict]:
"""Parses a PDF file using LlamaParse and extracts page texts and tables."""
# This requires `nest_asyncio.apply()` to be called once at the start of the app if running async.
# In a Streamlit app, ensure it's at the very top level if needed.
# import nest_asyncio; nest_asyncio.apply() # Uncomment if needed for async parser
parser = LlamaParse(
result_type="markdown",
skip_diagonal_text=True,
fast_mode=False,
num_workers=1, # Adjust as per environment capabilities
check_interval=10,
api_key=llamaparse_api_key
)
json_objs = parser.get_json_result(pdf_path)
page_texts, tables = {}, {}
for obj in json_objs:
json_list = obj['pages']
name = obj["file_path"].split("/")[-1]
page_texts[name] = {}
tables[name] = {}
for json_item in json_list:
for component in json_item['items']:
if component['type'] == 'table':
tables[name][json_item['page']] = component['rows']
return page_texts, tables
def generate_hypothetical_questions(llm: ChatOpenAI, docs: List[Document], is_table: bool = False) -> List[Document]:
"""Generates hypothetical questions for text chunks or tables."""
prompt_template = """
Generate a list of exactly three (3) hypothetical questions that the below nutritional disorder {content_type} could be used to answer:
{content}
Ensure that the questions are specific in the context of nutrition, dietary deficiencies, metabolic disorders, vitamin and mineral imbalances, obesity, and related health conditions.
Generate only a list of questions.
Do not mention anything before or after the list.
If the content cannot answer any questions, return an empty list.
"""
hyp_docs = []
content_type = "table" if is_table else "document"
for i, doc in enumerate(docs):
content_to_use = str(doc) if is_table else doc.page_content # Tables are often raw data, stringify
try:
response = llm.invoke(prompt_template.format(content_type=content_type, content=content_to_use))
questions = response.content
except Exception as e:
st.error(f"Error generating hypothetical questions for {'table' if is_table else 'text'} chunk ID {doc.id}: {e}")
questions = "[]" # Return empty list string on error
questions_metadata = {
'original_content': content_to_use,
'source': doc.metadata.get('source', 'unknown'),
'page': doc.metadata.get('page', -1),
'type': content_type
}
hyp_docs.append(
Document(
id=f"{'table_' if is_table else 'text_chunk_'}{doc.id or i}", # Ensure unique IDs
page_content=questions,
metadata=questions_metadata
)
)
time.sleep(0.1) # Small delay to avoid rate limits
return hyp_docs
def create_and_persist_vector_db(
documents: List[Document],
embedding_model: OpenAIEmbeddings,
collection_name: str,
persist_directory: str
):
"""Creates/updates a Chroma vector store and persists it."""
# Ensure IDs are strings as required by ChromaDB
doc_ids = [str(d.id) for d in documents] if documents else []
if not doc_ids:
st.warning(f"No documents to add to collection '{collection_name}'.")
return
# Initialize or connect to Chroma vectorstore
vector_store = Chroma.from_documents(
documents,
embedding_model,
collection_name=collection_name,
persist_directory=persist_directory
)
st.info(f"Initialized ChromaDB with collection '{collection_name}' at '{persist_directory}'. "
f"Documents count: {len(documents)}")
return vector_store
def load_vector_db(
embedding_model: OpenAIEmbeddings,
collection_name: str,
persist_directory: str
) -> Chroma:
"""Loads an existing Chroma vector store."""
try:
# Check if the directory exists and contains ChromaDB files
if not os.path.exists(persist_directory) or not os.listdir(persist_directory):
st.error(f"Vector DB directory '{persist_directory}' is empty or does not exist.") or print(f"Vector DB directory '{persist_directory}' is empty or does not exist.")
st.warning("Please ensure the 'nutritional_db' folder is correctly placed/mounted.") or print("Please ensure the 'nutritional_db' folder is correctly placed/mounted.")
return None
vector_store = Chroma(
collection_name=collection_name,
persist_directory=persist_directory,
embedding_function=embedding_model
)
st.success(f"Successfully loaded ChromaDB collection '{collection_name}' from '{persist_directory}'.") or print(f"Successfully loaded ChromaDB collection '{collection_name}' from '{persist_directory}'.")
# You can add a check for the number of documents loaded for verification
# Example: print(vector_store._collection.count())
return vector_store
except Exception as e:
st.error(f"Error loading ChromaDB from '{persist_directory}': {e}") or print(f"Error loading ChromaDB from '{persist_directory}': {e}")
return None
# --- 3. Agent Workflow Definition (LangGraph) ---
class AgentState(TypedDict):
"""Represents the state of the AI agent at different stages of the workflow."""
query: str
expanded_query: str
context: List[Dict[str, Any]]
response: str
precision_score: float
groundedness_score: float
groundedness_loop_count: int
precision_loop_count: int
feedback: str
query_feedback: str
groundedness_check: bool # This field isn't used in should_continue_groundedness, can be removed
loop_max_iter: int
# Node functions for LangGraph
def expand_query(state: AgentState) -> AgentState:
st.write("---Expanding Query---")
system_message = '''
You are a domain expert assisting in answering questions related to research papers.
Convert the user query into something that a nutritionist would understand. Use domain related words.
Return three (3) related search queries based on the user's request separated by newline.
Return only three (3) versions of the question as a list.
Perform query expansion on the question received. If there are multiple common ways of phrasing a user question \
or common synonyms for key words in the question, make sure to return multiple versions \
of the query with the different phrasings.
If the query has multiple parts, split them into separate simpler queries. This is the only case where you can generate more than three (3) queries.
If there are acronyms or words you are not familiar with, do not try to rephrase them.
Generate only a list of questions. Do not mention anything before or after the list.
Use the query feedback if provided to craft the search queries.
'''
expand_prompt = ChatPromptTemplate.from_messages([
("system", system_message),
("user", "Expand this query: {query} using the feedback: {query_feedback}")
])
chain = expand_prompt | st.session_state.llm | StrOutputParser() # Use llm from session state
expanded_query = chain.invoke({"query": state['query'], "query_feedback":state["query_feedback"]})
st.write(f"Expanded query:\n{expanded_query}")
state["expanded_query"] = expanded_query
return state
def retrieve_context(state: AgentState) -> AgentState:
st.write("---Retrieving Context---")
query = f"{state['query']}; {state['expanded_query']}"
st.write(f"Query used for retrieval:\n{query}")
# Ensure vector_store is loaded and available in session_state
if 'vector_store' not in st.session_state or st.session_state.vector_store is None:
st.error("Vector store not initialized. Cannot retrieve context.")
state['context'] = []
return state
retriever = st.session_state.vector_store.as_retriever(
search_type='similarity',
search_kwargs={'k': 3}
)
docs = retriever.invoke(query)
st.write(f"Retrieved documents (first 100 chars each):\n{[doc.page_content[:100] for doc in docs]}")
context = [
{"content": doc.page_content, "metadata": doc.metadata}
for doc in docs
]
state['context'] = context
#st.write(f"Extracted context with metadata:\n{context}") # Too verbose for production UI
return state
def craft_response(state: AgentState) -> AgentState:
st.write("---Crafting Response---")
system_message = '''
Generates a response to a user query and context provided.
Parameters:
query (str): The user's query and expanded queries based on user's query.
context (str): The documents retrieved relevant to the queries.
Returns:
response (str): The response generated by the model.
The function performs the following steps:
1. Constructs a prompt containing system and user prompts.
2. Sends the prompt containing user queries with context provided to the GPT model to generate a response.
3. Displays the response.
The answer you provide must come from the user queries with context provided.
If feedback is provided, use it to craft the response.
If information provided is not enough to answer the query respond with 'I don't know the answer. Not in my records.'
'''
response_prompt = ChatPromptTemplate.from_messages([
("system", system_message),
("user", "Query:\n{query}\nContext:\n{context}\n\nfeedback:\n{feedback}")
])
chain = response_prompt | st.session_state.llm # Use llm from session state
response = chain.invoke({
"query": state['query'],
"context": "\n".join([doc["content"] for doc in state['context']]),
"feedback": state["feedback"]
})
state['response'] = response.content # Access content from AIMessage
st.write(f"Intermediate response:\n{state['response']}")
return state
def score_groundedness(state: AgentState) -> AgentState:
st.write("---Checking Groundedness---")
system_message = '''
You are tasked with rating AI generated answers to questions posed by users.
Please act as an impartial judge and evaluate the quality of the provided answer which attempts to answer the provided question based on a provided context.
In the input, the context is {context}, while the AI generated response is {response}.
Evaluation criteria:
The task is to judge the extent to which the metric is followed by the answer.
1 - The metric is not followed at all
2 - The metric is followed only to a limited extent
3 - The metric is followed to a good extent
4 - The metric is followed mostly
5 - The metric is followed completely
The answer should be derived only from the information presented in the context
Do not show any instructions for deriving your answer.
Output your result as a float number between 0 and 1 using the evaluation criteria.
The better the criteria, the closer it is to 1 and the worse the criteria, the closer it is to 0.
'''
groundedness_prompt = ChatPromptTemplate.from_messages([
("system", system_message),
("user", "Context: {context}\nResponse: {response}\n\nGroundedness score:")
])
chain = groundedness_prompt | st.session_state.llm | StrOutputParser() # Use llm from session state
groundedness_score = float(chain.invoke({
"context": "\n".join([doc["content"] for doc in state['context']]),
"response": state['response']
}))
state['groundedness_score'] = groundedness_score
state['groundedness_loop_count'] += 1
st.write(f"Groundedness score: {groundedness_score}")
return state
def check_precision(state: AgentState) -> AgentState:
st.write("---Checking Precision---")
system_message = '''
Given question, answer and context verify if the context was useful in arriving at the given answer.
Give verdict as "1" if useful and "0" if not useful.
Output your result as a float number between 0 and 1
Give verdict as a scaled numeric value of type float between 0 and 1, such that
0 or near 0 if it is least useful, 0.5 or near 0.5 if retry is warranted, and 1 or close to 1 is most useful.
Do not show any instructions for deriving your answer.
'''
precision_prompt = ChatPromptTemplate.from_messages([
("system", system_message),
("user", "Query: {query}\nResponse: {response}\n\nPrecision score:")
])
chain = precision_prompt | st.session_state.llm | StrOutputParser() # Use llm from session state
precision_score = float(chain.invoke({
"query": state['query'],
"response": state['response']
}))
state['precision_score'] = precision_score
state['precision_loop_count'] +=1
st.write(f"Precision score: {precision_score}")
return state
def refine_response(state: AgentState) -> AgentState:
st.write("---Refining Response---")
system_message = '''
Since the last response failed the groundedness test, and is deemed not satisfactory,
use the feedback in terms of the query, context and the last response
to identify potential gaps, ambiguities, or missing details, and
to suggest improvements to enhance accuracy and completeness of the response.
'''
refine_response_prompt = ChatPromptTemplate.from_messages([
("system", system_message),
("user", "Query: {query}\nResponse: {response}\n\n"
"What improvements can be made to enhance accuracy and completeness?")
])
chain = refine_response_prompt | st.session_state.llm | StrOutputParser() # Use llm from session state
feedback = f"Previous Response: {state['response']}\nSuggestions: {chain.invoke({'query': state['query'], 'response': state['response']})}"
state['feedback'] = feedback
st.write(f"Refinement feedback:\n{feedback}")
return state
def refine_query(state: AgentState) -> AgentState:
st.write("---Refining Query---")
system_message = '''
Since the last response failed the precision test, and is deemed not satisfactory,
use the feedback in terms of the query, context and re-generate extended queries
to identify specific keywords, scope refinements, or missing details, and
to provides structured suggestions for improvement to enhance accuracy and completeness of the response.
'''
refine_query_prompt = ChatPromptTemplate.from_messages([
("system", system_message),
("user", "Original Query: {query}\nExpanded Query: {expanded_query}\n\n"
"What improvements can be made for a better search?")
])
chain = refine_query_prompt | st.session_state.llm | StrOutputParser() # Use llm from session state
query_feedback = f"Previous Expanded Query: {state['expanded_query']}\nSuggestions: {chain.invoke({'query': state['query'], 'expanded_query': state['expanded_query']})}"
state['query_feedback'] = query_feedback
st.write(f"Query refinement feedback:\n{query_feedback}")
return state
def should_continue_groundedness(state: AgentState) -> str:
st.write("---Deciding Groundedness Continuation---")
st.write(f"Groundedness loop count: {state['groundedness_loop_count']}")
if state['groundedness_score'] >= 0.8:
st.write("Moving to precision check.")
return "check_precision"
else:
if state["groundedness_loop_count"] >= state['loop_max_iter']:
st.write("Max iterations reached for groundedness.")
return "max_iterations_reached"
else:
st.write("Groundedness score not met. Refining response.")
return "refine_response"
def should_continue_precision(state: AgentState) -> str:
st.write("---Deciding Precision Continuation---")
st.write(f"Precision loop count: {state['precision_loop_count']}")
if state['precision_score'] > 0.8:
st.write("Precision sufficient. Ending workflow.")
return "pass"
else:
if state["precision_loop_count"] >= state['loop_max_iter']:
st.write("Max iterations reached for precision.")
return "max_iterations_reached"
else:
st.write("Precision score not met. Refining query.")
return "refine_query"
def max_iterations_reached(state: AgentState) -> AgentState:
st.write("---Max Iterations Reached---")
response = "I'm unable to refine the response further. Please provide more context or clarify your question."
state['response'] = response
return state
@tool
def agentic_rag_tool(query: str) -> Dict[str, Any]:
"""
Runs the RAG-based agent workflow to generate context-aware responses.
This function is exposed as a tool for the overall chatbot.
"""
# Initialize state for the LangGraph workflow
inputs = {
"query": query,
"expanded_query": "",
"context": [],
"response": "",
"precision_score": 0.0,
"groundedness_score": 0.0,
"groundedness_loop_count": 0,
"precision_loop_count": 0,
"feedback": "",
"query_feedback": "",
"loop_max_iter": 3
}
# Compile the workflow once and store it in session state if not already done
if 'workflow_app' not in st.session_state:
st.session_state.workflow_app = create_rag_workflow().compile()
# Invoke the compiled LangGraph workflow
output = st.session_state.workflow_app.invoke(inputs)
return output
def create_rag_workflow() -> StateGraph:
"""Creates the LangGraph workflow for the RAG agent."""
workflow = StateGraph(AgentState)
workflow.add_node("expand_query", expand_query)
workflow.add_node("retrieve_context", retrieve_context)
workflow.add_node("craft_response", craft_response)
workflow.add_node("score_groundedness", score_groundedness)
workflow.add_node("refine_response", refine_response)
workflow.add_node("check_precision", check_precision)
workflow.add_node("refine_query", refine_query)
workflow.add_node("max_iterations_reached", max_iterations_reached)
workflow.add_edge(START, "expand_query")
workflow.add_edge("expand_query", "retrieve_context")
workflow.add_edge("retrieve_context", "craft_response")
workflow.add_edge("craft_response", "score_groundedness")
workflow.add_conditional_edges(
"score_groundedness",
should_continue_groundedness,
{
"check_precision": "check_precision",
"refine_response": "refine_response",
"max_iterations_reached": "max_iterations_reached"
}
)
workflow.add_edge("refine_response", "craft_response")
workflow.add_conditional_edges(
"check_precision",
should_continue_precision,
{
"pass": END,
"refine_query": "refine_query",
"max_iterations_reached": "max_iterations_reached"
}
)
workflow.add_edge("refine_query", "expand_query")
workflow.add_edge("max_iterations_reached", END)
return workflow
# --- 4. Main Chatbot Class (Integrating Memory & Agent) ---
class NutritionBot:
def __init__(self, config: Dict):
"""
Initialize the NutritionBot class, setting up memory, the LLM client, tools, and the agent executor.
"""
mem0_api_key = config["MEM0_API_KEY"]
openai_api_key = config["AZURE_OPENAI_API_KEY"]
openai_api_base = config["AZURE_OPENAI_API_BASE"]
# Initialize a memory client to store and retrieve customer interactions
self.memory = MemoryClient(api_key=mem0_api_key)
# Initialize the OpenAI client (LangChain ChatOpenAI)
self.client = ChatOpenAI(
model="gpt-4o-mini",
openai_api_key=openai_api_key,
openai_api_base=openai_api_base,
temperature=0
)
# Store LLM in session state for use in graph nodes
st.session_state.llm = self.client
# Define tools available to the chatbot
tools = [agentic_rag_tool]
# Define the system prompt for the agent
system_prompt = """You are a caring and knowledgeable Medical Support Agent, specializing in nutrition disorder-related guidance. Your goal is to provide accurate, empathetic, and tailored nutritional recommendations while ensuring a seamless customer experience.
Guidelines for Interaction:
Maintain a polite, professional, and reassuring tone.
Show genuine empathy for customer concerns and health challenges.
Reference past interactions to provide personalized and consistent advice.
Engage with the customer by asking about their food preferences, dietary restrictions, and lifestyle before offering recommendations.
Ensure consistent and accurate information across conversations.
If any detail is unclear or missing, proactively ask for clarification.
Always use the agentic_rag_tool to retrieve up-to-date and evidence-based nutrition insights.
Keep track of ongoing issues and follow-ups to ensure continuity in support.
Your primary goal is to help customers make informed nutrition decisions that align with their health conditions and personal preferences.
"""
# Build the prompt template for the agent
prompt = ChatPromptTemplate.from_messages([
("system", system_prompt),
("human", "{input}"),
("placeholder", "{agent_scratchpad}")
])
# Create an agent capable of interacting with tools and executing tasks
agent = create_tool_calling_agent(self.client, tools, prompt)
# Wrap the agent in an executor to manage tool interactions and execution flow
self.agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)
def store_customer_interaction(self, user_id: str, message: str, response: str, metadata: Dict = None):
"""Store customer interaction in memory for future reference."""
if metadata is None:
metadata = {}
metadata["timestamp"] = datetime.now().isoformat()
conversation = [
{"role": "user", "content": message},
{"role": "assistant", "content": response}
]
self.memory.add(conversation, user_id=user_id, output_format="v1.1", metadata=metadata)
def get_relevant_history(self, user_id: str, query: str) -> List[Dict]:
"""Retrieve past interactions relevant to the current query."""
return self.memory.search(query=query, user_id=user_id, limit=5)
def handle_customer_query(self, user_id: str, query: str) -> str:
"""Process a customer's query and provide a response, taking into account past interactions."""
relevant_history = self.get_relevant_history(user_id, query)
context = "Previous relevant interactions:\n"
for memory_item in relevant_history:
# Mem0 'memory' field is typically a list of dicts or a string.
# Assuming 'v1.1' output format from `memory.add` means `memory_item['memory']` is structured.
if isinstance(memory_item.get('memory'), list):
for part in memory_item['memory']:
context += f"{part['role'].capitalize()}: {part['content']}\n"
else: # Fallback for older formats or if it's a simple string
context += f"History: {memory_item.get('memory', 'N/A')}\n"
context += "---\n"
prompt = f"""
Context:
{context}
Current customer query: {query}
Provide a helpful response that takes into account any relevant past interactions.
"""
st.write("Prompt sent to agent executor:", prompt) # Debugging
try:
response_dict = self.agent_executor.invoke({"input": prompt})
response_content = response_dict.get('output', "I'm sorry, I couldn't generate a response.")
except Exception as e:
st.error(f"Error during agent execution: {e}")
response_content = f"I'm sorry, I encountered an internal error: {e}"
self.store_customer_interaction(user_id=user_id, message=query, response=response_content, metadata={"type": "support_query"})
return response_content
# --- 5. Streamlit UI ---
def nutrition_disorder_streamlit_app():
"""Streamlit-based UI for the Nutrition Disorder Specialist Agent."""
st.set_page_config(page_title="Nutrition Disorder Specialist", layout="centered")
st.title("👨‍⚕️ Nutrition Disorder Specialist")
st.markdown("Ask me anything about nutrition disorders, symptoms, causes, treatments, and more.")
st.markdown("---")
# Initialize session state variables
if 'chat_history' not in st.session_state:
st.session_state.chat_history = []
if 'user_id' not in st.session_state:
st.session_state.user_id = None
if 'chatbot' not in st.session_state:
st.session_state.chatbot = None
if 'config_loaded' not in st.session_state:
st.session_state.config_loaded = False
if 'vector_store_loaded' not in st.session_state:
st.session_state.vector_store_loaded = False
# --- Configuration Loading and Model Initialization ---
if not st.session_state.config_loaded:
with st.spinner("Loading configurations and initializing models..."):
config = load_config_from_env()
if not all(config.values()):
st.error("Some environment variables are missing. Please set them up for the app to function.")
st.stop() # Stop execution if critical configs are missing
# Step 1.
embedding_model, llm_instance, embedding_function, llama_guard_client_instance = initialize_llms_and_embeddings(config)
# Step 2. Store initialized objects in session state
st.session_state.config = config
st.session_state.embedding_model = embedding_model
st.session_state.llm = llm_instance
st.session_state.embedding_function = embedding_function # Used during vector_store creation/loading
st.session_state.llama_guard_client = llama_guard_client_instance
st.session_state.config_loaded = True
st.rerun() # Rerun to update UI after loading
# --- Vector Store Loading ---
if st.session_state.config_loaded and not st.session_state.vector_store_loaded:
with st.spinner("Loading nutrition knowledge base (vector database)..."):
# Ensure the nutritional_db directory exists relative to the app.py
# In Docker, this means the folder should be copied into /app
persist_dir = "./nutritional_db"
if not os.path.exists(persist_dir):
st.error(f"Required data directory '{persist_dir}' not found. Please ensure it's copied into the Docker image.")
st.stop()
st.session_state.vector_store = load_vector_db(
embedding_model=st.session_state.embedding_model,
collection_name="nutritional_hypotheticals",
persist_directory=persist_dir
)
if st.session_state.vector_store is None:
st.error("Failed to load vector database. Chat functionality will be limited.")
st.session_state.vector_store_loaded = True
st.rerun() # Rerun to update UI after loading
# --- Login Form ---
if st.session_state.user_id is None:
with st.form("login_form", clear_on_submit=True):
user_id_input = st.text_input("Please enter your name to begin:", key="user_id_input")
submit_button = st.form_submit_button("Login")
if submit_button and user_id_input:
st.session_state.user_id = user_id_input.strip()
st.session_state.chat_history.append({
"role": "assistant",
"content": f"Welcome, {st.session_state.user_id}! How can I help you with nutrition disorders today?"
})
# Initialize chatbot only after config and vector store are ready
if st.session_state.config_loaded and st.session_state.vector_store_loaded:
st.session_state.chatbot = NutritionBot(st.session_state.config)
else:
st.warning("Chatbot initialization delayed as configurations or vector store are still loading.")
st.rerun()
# --- Chat Interface ---
elif st.session_state.chatbot: # Only show chat if chatbot is initialized
# Display chat history
for message in st.session_state.chat_history:
with st.chat_message(message["role"]):
st.write(message["content"])
user_query = st.chat_input("Type your question here (e.g., 'What are dietary deficiencies?')")
if user_query:
if user_query.lower() == "exit":
st.session_state.chat_history.append({"role": "user", "content": "exit"})
with st.chat_message("user"):
st.write("exit")
goodbye_msg = "Goodbye! Feel free to return if you have more questions about nutrition disorders."
st.session_state.chat_history.append({"role": "assistant", "content": goodbye_msg})
with st.chat_message("assistant"):
st.write(goodbye_msg)
st.session_state.user_id = None # Log out
st.session_state.chatbot = None # Clear chatbot instance
st.session_state.chat_history = [] # Clear history on exit
st.rerun()
return
st.session_state.chat_history.append({"role": "user", "content": user_query})
with st.chat_message("user"):
st.write(user_query)
# Filter input using Llama Guard
with st.spinner("Filtering input for safety..."):
filtered_result = filter_input_with_llama_guard(user_query, st.session_state.llama_guard_client)
if filtered_result:
filtered_result = filtered_result.replace("\n", " ").strip()
st.info(f"Llama Guard says: {filtered_result}") # Show Llama Guard's verdict
# Process the user query if safe
if filtered_result and ("safe" in filtered_result.lower() or "s7" in filtered_result.lower()): # Allow "safe S7" etc.
with st.spinner("Thinking..."):
try:
response = st.session_state.chatbot.handle_customer_query(st.session_state.user_id, user_query)
st.session_state.chat_history.append({"role": "assistant", "content": response})
with st.chat_message("assistant"):
st.write(response)
except Exception as e:
error_msg = f"Sorry, I encountered an error while processing your query. Please try again. Error: {e}"
st.error(error_msg)
st.session_state.chat_history.append({"role": "assistant", "content": error_msg})
else:
inappropriate_msg = "I apologize, but I cannot process that input as it may be inappropriate or unsafe. Please try again."
st.session_state.chat_history.append({"role": "assistant", "content": inappropriate_msg})
with st.chat_message("assistant"):
st.write(inappropriate_msg)
st.rerun() # Rerun to update chat history instantly
elif st.session_state.user_id: # User is logged in but chatbot not ready
st.info("Initializing chatbot. Please wait...")
# --- Main entry point for Streamlit App ---
if __name__ == "__main__":
nutrition_disorder_streamlit_app()