ChatBotAgenticRAG1 / pipeline.py
Phoenix21's picture
Update pipeline.py
ac8126f verified
import os
import getpass
import spacy
import pandas as pd
import numpy as np
from typing import Optional, List, Dict, Any
import subprocess
from langchain.llms.base import LLM
from langchain.docstore.document import Document
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.vectorstores import FAISS
from langchain.chains import RetrievalQA
from smolagents import DuckDuckGoSearchTool, ManagedAgent, LiteLLMModel ,CodeAgent, HfApiModel
from pydantic import BaseModel, Field, ValidationError, validator
from mistralai import Mistral
# Import Google Gemini model
from langchain_google_genai import ChatGoogleGenerativeAI
from classification_chain import get_classification_chain
from cleaner_chain import get_cleaner_chain
from refusal_chain import get_refusal_chain
from tailor_chain import get_tailor_chain
from prompts import classification_prompt, refusal_prompt, tailor_prompt
LANGSMITH_TRACING=True
LANGSMITH_ENDPOINT="https://api.smith.langchain.com"
LANGSMITH_API_KEY=os.environ.get("LANGSMITH_API_KEY")
LANGSMITH_PROJECT=os.environ.get("LANGCHAIN_PROJECT")
# Initialize Mistral API client
mistral_api_key = os.environ.get("MISTRAL_API_KEY")
client = Mistral(api_key=mistral_api_key)
# Setup ChatGoogleGenerativeAI for Gemini
# Ensure GEMINI_API_KEY is set in your environment variables.
gemini_llm = ChatGoogleGenerativeAI(
model="gemini-1.5-pro",
temperature=0.5,
max_retries=2,
google_api_key=os.environ.get("GEMINI_API_KEY"),
# Additional parameters or safety_settings can be added here if needed
)
# web_gemini_llm = LiteLLMModel(model_id="gemini/gemini-pro", api_key=os.environ.get("GEMINI_API_KEY"))
################################################################################
# Pydantic Models
################################################################################
class QueryInput(BaseModel):
query: str = Field(..., min_length=1, description="The input query string")
@validator('query')
def check_query_is_string(cls, v):
if not isinstance(v, str):
raise ValueError("Query must be a valid string")
if v.strip() == "":
raise ValueError("Query cannot be empty or just whitespace")
return v.strip()
class ModerationResult(BaseModel):
is_safe: bool = Field(..., description="Whether the content is safe")
categories: Dict[str, bool] = Field(default_factory=dict, description="Detected content categories")
original_text: str = Field(..., description="The original input text")
################################################################################
# SPACy Setup
################################################################################
def install_spacy_model():
try:
spacy.load("en_core_web_sm")
print("spaCy model 'en_core_web_sm' is already installed.")
except OSError:
print("Downloading spaCy model 'en_core_web_sm'...")
subprocess.run(["python", "-m", "spacy", "download", "en_core_web_sm"], check=True)
print("spaCy model 'en_core_web_sm' downloaded successfully.")
install_spacy_model()
nlp = spacy.load("en_core_web_sm")
################################################################################
# Utility Functions
################################################################################
def sanitize_message(message: Any) -> str:
"""Sanitize message input to ensure it's a valid string."""
try:
if hasattr(message, 'content'):
return str(message.content).strip()
if isinstance(message, dict) and 'content' in message:
return str(message['content']).strip()
if isinstance(message, list) and len(message) > 0:
if isinstance(message[0], dict) and 'content' in message[0]:
return str(message[0]['content']).strip()
if hasattr(message[0], 'content'):
return str(message[0].content).strip()
return str(message).strip()
except Exception as e:
raise RuntimeError(f"Error in sanitize function: {str(e)}")
def extract_main_topic(query: str) -> str:
"""Extracts a main topic (named entity or noun) from the user query."""
try:
query_input = QueryInput(query=query)
doc = nlp(query_input.query)
main_topic = None
# Attempt to find an entity
for ent in doc.ents:
if ent.label_ in ["ORG", "PRODUCT", "PERSON", "GPE", "TIME"]:
main_topic = ent.text
break
# If no named entity, fall back to nouns or proper nouns
if not main_topic:
for token in doc:
if token.pos_ in ["NOUN", "PROPN"]:
main_topic = token.text
break
return main_topic if main_topic else "this topic"
except Exception as e:
print(f"Error extracting main topic: {e}")
return "this topic"
def moderate_text(query: str) -> ModerationResult:
"""Uses Mistral's moderation to determine if the content is safe."""
try:
query_input = QueryInput(query=query)
response = client.classifiers.moderate_chat(
model="mistral-moderation-latest",
inputs=[{"role": "user", "content": query_input.query}]
)
is_safe = True
categories = {}
if hasattr(response, 'results') and response.results:
categories = {
"violence": response.results[0].categories.get("violence_and_threats", False),
"hate": response.results[0].categories.get("hate_and_discrimination", False),
"dangerous": response.results[0].categories.get("dangerous_and_criminal_content", False),
"selfharm": response.results[0].categories.get("selfharm", False)
}
# If any flagged category is True, then not safe
is_safe = not any(categories.values())
return ModerationResult(
is_safe=is_safe,
categories=categories,
original_text=query_input.query
)
except ValidationError as e:
raise ValueError(f"Input validation failed: {str(e)}")
except Exception as e:
raise RuntimeError(f"Moderation failed: {str(e)}")
def classify_query(query: str) -> str:
"""Classify user query into known categories using your classification chain."""
try:
query_input = QueryInput(query=query)
# Quick pattern-based approach for 'Wellness'
# wellness_keywords = ["box breathing", "meditation", "yoga", "mindfulness", "breathing exercises"]
wellness_keywords=[]
if any(keyword in query_input.query.lower() for keyword in wellness_keywords):
return "Wellness"
# Use chain for everything else
class_result = classification_chain.invoke({"query": query_input.query})
print(class_result)
# classification = class_result.get("text", "").strip()
classification=class_result
return classification if classification != "" else "OutOfScope"
except ValidationError as e:
raise ValueError(f"Classification input validation failed: {str(e)}")
except Exception as e:
raise RuntimeError(f"Classification failed: {str(e)}")
################################################################################
# Vector Store Building/Loading
################################################################################
def build_or_load_vectorstore(csv_path: str, store_dir: str) -> FAISS:
try:
if os.path.exists(store_dir):
print(f"DEBUG: Found existing FAISS store at '{store_dir}'. Loading...")
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/multi-qa-mpnet-base-dot-v1")
vectorstore = FAISS.load_local(store_dir, embeddings)
return vectorstore
else:
print(f"DEBUG: Building new store from CSV: {csv_path}")
df = pd.read_csv(csv_path)
df = df.loc[:, ~df.columns.str.contains('^Unnamed')]
df.columns = df.columns.str.strip()
if "Answer" in df.columns:
df.rename(columns={"Answer": "Answers"}, inplace=True)
if "Question" not in df.columns and "Question " in df.columns:
df.rename(columns={"Question ": "Question"}, inplace=True)
if "Question" not in df.columns or "Answers" not in df.columns:
raise ValueError("CSV must have 'Question' and 'Answers' columns.")
docs = []
for _, row in df.iterrows():
q = str(row["Question"])
ans = str(row["Answers"])
doc = Document(page_content=ans, metadata={"question": q})
docs.append(doc)
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/multi-qa-mpnet-base-dot-v1")
vectorstore = FAISS.from_documents(docs, embedding=embeddings)
vectorstore.save_local(store_dir)
return vectorstore
except Exception as e:
raise RuntimeError(f"Error building/loading vector store: {str(e)}")
def build_rag_chain(vectorstore: FAISS) -> RetrievalQA:
"""Build RAG chain using the Gemini LLM directly without a custom class."""
try:
retriever = vectorstore.as_retriever(search_type="similarity", search_kwargs={"k": 3})
chain = RetrievalQA.from_chain_type(
llm=gemini_llm, # Directly use the ChatGoogleGenerativeAI instance
chain_type="stuff",
retriever=retriever,
return_source_documents=True
)
return chain
except Exception as e:
raise RuntimeError(f"Error building RAG chain: {str(e)}")
################################################################################
# Web Search Caching: Separate FAISS Vector Store
################################################################################
# Directory for storing cached web search results
web_search_store_dir = "faiss_websearch_store"
def build_or_load_websearch_store(store_dir: str) -> FAISS:
"""
Builds or loads a FAISS vector store for caching web search results.
Each Document will have page_content as the search result text,
and metadata={"question": <user_query>}.
"""
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/multi-qa-mpnet-base-dot-v1")
if os.path.exists(store_dir):
print(f"DEBUG: Found existing WebSearch FAISS store at '{store_dir}'. Loading...")
return FAISS.load_local(store_dir, embeddings)
else:
print(f"DEBUG: Creating a new, empty WebSearch FAISS store at '{store_dir}'...")
# Start empty
empty_store = FAISS.from_texts([""], embeddings, metadatas=[{"question": "placeholder"}])
# Remove the placeholder doc so we don't retrieve it
empty_store.index.reset()
empty_store.docstore._dict = {}
empty_store.save_local(store_dir)
return empty_store
# Initialize the web search vector store
web_search_vectorstore = build_or_load_websearch_store(web_search_store_dir)
websearch_embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/multi-qa-mpnet-base-dot-v1")
def compute_cosine_similarity(vec_a: List[float], vec_b: List[float]) -> float:
"""Compute cosine similarity between two embedding vectors."""
a = np.array(vec_a, dtype=float)
b = np.array(vec_b, dtype=float)
return float(np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b) + 1e-10))
def get_cached_websearch(query: str, threshold: float = 0.8) -> Optional[str]:
"""
Attempts to retrieve a cached web search result for a given query.
If the top retrieved document has a cosine similarity >= threshold,
returns that document's page_content. Otherwise, returns None.
"""
# Retrieve the top doc from the store
retriever = web_search_vectorstore.as_retriever(search_type="similarity", search_kwargs={"k": 1})
results = retriever.get_relevant_documents(query)
if not results:
return None
# Compare similarity with the top doc
top_doc = results[0]
query_vec = websearch_embeddings.embed_query(query)
doc_vec = websearch_embeddings.embed_query(top_doc.page_content)
similarity = compute_cosine_similarity(query_vec, doc_vec)
if similarity >= threshold:
print(f"DEBUG: Using cached web search (similarity={similarity:.2f} >= {threshold})")
return top_doc.page_content
print(f"DEBUG: Cached doc similarity={similarity:.2f} < {threshold}, not reusing.")
return None
def store_websearch_result(query: str, web_search_text: str):
"""
Embeds and stores the web search result text in the web search vector store,
keyed by the question in metadata. Then saves the store locally.
"""
if not web_search_text.strip():
return # Don't store empty results
doc = Document(page_content=web_search_text, metadata={"question": query})
web_search_vectorstore.add_documents([doc], embedding=websearch_embeddings)
web_search_vectorstore.save_local(web_search_store_dir)
def do_cached_web_search(query: str) -> str:
"""Perform a DuckDuckGo web search, but with caching via FAISS vector store."""
# 1) Check cache
cached_result = get_cached_websearch(query)
if cached_result:
return cached_result
# 2) If no suitable cached answer, do a new search
try:
print("DEBUG: Performing a new web search...")
# model = LiteLLMModel(model_id="gemini/gemini-pro", api_key=os.environ.get("GEMINI_API_KEY"))
model=HfApiModel()
search_tool = DuckDuckGoSearchTool()
web_agent = CodeAgent(
tools=[search_tool],
model=model
)
managed_web_agent = ManagedAgent(
agent=web_agent,
name="web_search",
description="Runs a web search for you. Provide your query as an argument."
)
manager_agent = CodeAgent(
tools=[], # If you have additional tools for the manager, add them here
model=model,
managed_agents=[managed_web_agent]
)
new_search_result = manager_agent.run(f"Search for information about: {query}")
# 3) Store in cache for future reuse
store_websearch_result(query, new_search_result)
return str(new_search_result).strip()
except Exception as e:
print(f"Web search failed: {e}")
return ""
################################################################################
# Response Merging
################################################################################
def merge_responses(csv_answer: str, web_answer: str) -> str:
"""Merge CSV-based RAG result with web search results."""
try:
if not csv_answer and not web_answer:
return "I apologize, but I couldn't find any relevant information."
if not web_answer:
return csv_answer
if not csv_answer:
return web_answer
return f"{csv_answer}\n\nAdditional information from web search:\n{web_answer}"
except Exception as e:
print(f"Error merging responses: {e}")
return csv_answer or web_answer or "I apologize, but I couldn't process the information properly."
################################################################################
# Main Pipeline
################################################################################
def run_pipeline(query: str) -> str:
"""
Pipeline logic to:
1) Sanitize & moderate the query
2) Classify the query (OutOfScope, Wellness, Brand, etc.)
3) If safe & in scope, do RAG + ALWAYS do a cached web search
4) Merge responses and tailor final output
"""
try:
print(query)
sanitized_query = sanitize_message(query)
query_input = QueryInput(query=sanitized_query)
topic = extract_main_topic(query_input.query)
moderation_result = moderate_text(query_input.query)
# Check for unsafe content
if not moderation_result.is_safe:
return "Sorry, this query contains harmful or inappropriate content."
# Classify
classification = classify_query(moderation_result.original_text)
# If out-of-scope, refuse
if classification == "OutOfScope":
refusal_text = refusal_chain.invoke({"topic": topic,"query":query})
return tailor_chain.run({"response": refusal_text}).strip()
# Otherwise, do a RAG query and also do a web search (cached)
if classification == "Wellness":
# RAG from wellness store
rag_result = wellness_rag_chain({"query": moderation_result.original_text})
csv_answer = rag_result.get("result", "").strip() if isinstance(rag_result, dict) else str(rag_result).strip()
# Always do a (cached) web search
web_answer = do_cached_web_search(moderation_result.original_text)
# Merge CSV & Web
final_merged = merge_responses(csv_answer, web_answer)
return tailor_chain.run({"response": final_merged}).strip()
if classification == "Brand":
# RAG from brand store
rag_result = brand_rag_chain({"query": moderation_result.original_text})
csv_answer = rag_result.get("result", "").strip() if isinstance(rag_result, dict) else str(rag_result).strip()
# Always do a (cached) web search
web_answer = do_cached_web_search(moderation_result.original_text)
# Merge CSV & Web
final_merged = merge_responses(csv_answer, web_answer)
return tailor_chain.run({"response": final_merged}).strip()
# If it doesn't fall under known categories, return refusal by default.
refusal_text = refusal_chain.invoke({"topic": topic,"query":query})
return tailor_chain.run({"response": refusal_text}).strip()
except ValidationError as e:
raise ValueError(f"Input validation failed: {str(e)}")
except Exception as e:
raise RuntimeError(f"Error in run_pipeline: {str(e)}")
def run_with_chain(query: str) -> str:
"""Convenience function to run the main pipeline and handle errors gracefully."""
try:
return run_pipeline(query)
except Exception as e:
print(f"Error in run_with_chain: {str(e)}")
return "I apologize, but I encountered an error processing your request. Please try again."
################################################################################
# Chain & Vectorstore Initialization
################################################################################
# Load your classification/refusal/tailor/cleaner chains
classification_chain = get_classification_chain()
refusal_chain = get_refusal_chain()
tailor_chain = get_tailor_chain()
cleaner_chain = get_cleaner_chain()
# CSV file paths and store directories for RAG
wellness_csv = "AIChatbot.csv"
brand_csv = "BrandAI.csv"
wellness_store_dir = "faiss_wellness_store"
brand_store_dir = "faiss_brand_store"
# Build or load the vector stores
wellness_vectorstore = build_or_load_vectorstore(wellness_csv, wellness_store_dir)
brand_vectorstore = build_or_load_vectorstore(brand_csv, brand_store_dir)
# Build RAG chains
wellness_rag_chain = build_rag_chain(wellness_vectorstore)
brand_rag_chain = build_rag_chain(brand_vectorstore)
print("Pipeline initialized successfully! Ready to handle querie with caching.")