final

chatbot.py

@@ -19,14 +19,15 @@ from peft import PeftModel, PeftConfig
 from sentence_transformers import SentenceTransformer
 
 # LangChain imports
-from langchain.llms import HuggingFacePipeline
-from langchain.chains import LLMChain
-from langchain.memory import ConversationBufferMemory
-from langchain.prompts import PromptTemplate
-from langchain.embeddings import HuggingFaceEmbeddings
-from langchain.text_splitter import RecursiveCharacterTextSplitter
-from langchain.document_loaders import TextLoader
-from langchain.vectorstores import FAISS
+# Core LangChain components for building conversational AI
+from langchain.llms import HuggingFacePipeline  # Wrapper for HuggingFace models
+from langchain.chains import LLMChain  # Chain for LLM interactions
+from langchain.memory import ConversationBufferMemory  # Memory for conversation history
+from langchain.prompts import PromptTemplate  # Template for structured prompts
+from langchain.embeddings import HuggingFaceEmbeddings  # Text embeddings for similarity search
+from langchain.text_splitter import RecursiveCharacterTextSplitter  # Document chunking
+from langchain.document_loaders import TextLoader  # Load text documents
+from langchain.vectorstores import FAISS  # Vector database for similarity search
 
 # Import FlowManager
 from conversation_flow import FlowManager
@@ -106,11 +107,8 @@ class SessionSummary(BaseModel):
     user_id: str = Field(
         ...,
         description="Identifier of the user",
-        examples=["user_123"]
-    )
-    start_time: str = Field(
-        ...,
-        description="ISO format start time of the session"
+        examples=["user_123"])
+    start_time: str = Field(..., description="ISO format start time of the session"
     )
     end_time: str = Field(
         ...,
@@ -266,12 +264,24 @@ class MentalHealthChatbot:
         self.flow_manager = FlowManager(self.llm)
         
         # Setup conversation memory with LangChain
+        # ConversationBufferMemory stores the conversation history in a buffer
+        # This allows the chatbot to maintain context across multiple interactions
+        # - return_messages=True: Returns messages as a list of message objects
+        # - input_key="input": Specifies which key to use for the input in the memory
         self.memory = ConversationBufferMemory(
             return_messages=True,
             input_key="input"
         )
         
         # Create conversation prompt template
+        # PromptTemplate defines the structure for generating responses
+        # It includes placeholders for dynamic content that gets filled during generation
+        # Input variables:
+        # - history: Previous conversation context from memory
+        # - input: Current user message
+        # - past_context: Relevant past conversations from vector search
+        # - emotion_context: Detected emotions and their context
+        # - guidelines: Relevant therapeutic guidelines from vector search
         self.prompt_template = PromptTemplate(
             input_variables=["history", "input", "past_context", "emotion_context", "guidelines"],
             template="""You are a supportive and empathetic mental health conversational AI. Your role is to provide therapeutic support while maintaining professional boundaries.
@@ -300,6 +310,12 @@ Response:"""
         )
         
         # Create the conversation chain
+        # LLMChain combines the language model, prompt template, and memory
+        # This creates a conversational agent that can:
+        # - Generate responses using the LLM
+        # - Use the prompt template for structured input
+        # - Maintain conversation history in memory
+        # - verbose=False: Disables detailed logging of chain operations
         self.conversation = LLMChain(
             llm=self.llm,
             prompt=self.prompt_template,
@@ -308,11 +324,16 @@ Response:"""
         )
         
         # Setup embeddings for vector search
+        # HuggingFaceEmbeddings converts text to numerical vectors for similarity search
+        # all-MiniLM-L6-v2 is a lightweight but effective sentence embedding model
+        # These embeddings enable semantic search of past conversations and guidelines
         self.embeddings = HuggingFaceEmbeddings(
             model_name="sentence-transformers/all-MiniLM-L6-v2"
         )
         
         # Setup vector database for retrieving relevant past conversations
+        # The vector database stores embeddings of therapy guidelines and past conversations
+        # This enables semantic search to find relevant context for each response
         if therapy_guidelines_path and os.path.exists(therapy_guidelines_path):
             self.setup_vector_db(therapy_guidelines_path)
         else:
@@ -425,6 +446,9 @@ Response:"""
             )
             
             # Create LangChain wrapper
+            # HuggingFacePipeline wraps the HuggingFace pipeline for use with LangChain
+            # This enables the pipeline to work seamlessly with LangChain components
+            # like chains, memory, and prompts
             llm = HuggingFacePipeline(pipeline=text_generator)
             
             return model, tokenizer, llm
@@ -579,19 +603,7 @@ Response:"""
                 response = response.split("Response:")[-1].strip()
                 response = re.sub(r'^(Hey|Hi|Hello|Hi there|Hey there),\s*', '', response)
         
-        # Limit response length
-        max_words = 60
-        max_sentences = 4
-
-        # Split into sentences
-        sentences = re.split(r'(?<=[.!?]) +', response)
-        if len(sentences) > max_sentences:
-            response = ' '.join(sentences[:max_sentences])
 
-        # Split into words
-        words = response.split()
-        if len(words) > max_words:
-            response = ' '.join(words[:max_words]) + '...'
         
         return response.strip()