add relevant content + modified prompts

chatbot.py

@@ -1,4 +1,3 @@
-
 import os
 from pydantic import Field
 from langchain_community.retrievers import PineconeHybridSearchRetriever
@@ -18,14 +17,9 @@ from langchain_core.runnables.history import RunnableWithMessageHistory
 from langchain.chains import create_history_aware_retriever
 from langchain.chains.combine_documents import create_stuff_documents_chain
 from dotenv import load_dotenv
-
-# --- New Imports for Reranking ---
 from langchain.retrievers.contextual_compression import ContextualCompressionRetriever
 from langchain_community.document_compressors import FlashrankRerank
-# It's also good practice to import Ranker from the FlashRank library directly
-# to avoid potential Pydantic errors
 from flashrank import Ranker
-# ----------------------------------
 
 load_dotenv()
 
@@ -46,8 +40,6 @@ class FixedDimensionGoogleGenerativeAIEmbeddings(GoogleGenerativeAIEmbeddings):
         None, description="The fixed output dimension for embeddings."
     )
 
-    # We override the __init__ to handle the parameter and pass it to the base class.
-    # The Field definition above will handle the validation, so we don't need a custom pop.
     def __init__(self, **kwargs):
         super().__init__(**kwargs)
 
@@ -61,8 +53,6 @@ class FixedDimensionGoogleGenerativeAIEmbeddings(GoogleGenerativeAIEmbeddings):
             kwargs['output_dimensionality'] = self.output_dimensionality
         return super().embed_query(text, **kwargs)
 
-# Now, you can use your new class as intended
-# You can pass the output_dimensionality to the constructor directly.
 embeddings = FixedDimensionGoogleGenerativeAIEmbeddings(
     google_api_key=GOOGLE_API_KEY,
     model=embed_model,
@@ -115,10 +105,6 @@ class CustomHybridSearchRetriever(PineconeHybridSearchRetriever):
             docs.append(doc)
         return docs
 
-# --- New Reranker Integration Section ---
-
-# 1. Update the top_k for your base retriever to fetch more documents.
-# We will fetch a larger set (e.g., top-50) to give the reranker more options.
 namespace = 'portfolio'
 base_retriever = CustomHybridSearchRetriever( 
     embeddings=embeddings, 
@@ -128,25 +114,20 @@ base_retriever = CustomHybridSearchRetriever(
     namespace=namespace
 )
 
-# 2. Define the FlashRank reranker (the "compressor").
-# We specify the top_n to return after reranking (e.g., top 5).
 reranker_compressor = FlashrankRerank(
-    model=rerank_model, # The default lightweight model
-    top_n=5 # Return the top 5 most relevant documents
+    model=rerank_model,
+    top_n=5 
 )
 
-# 3. Create the ContextualCompressionRetriever.
-# This wraps your base hybrid search retriever and applies the reranker.
 retriever = ContextualCompressionRetriever(
     base_compressor=reranker_compressor, 
     base_retriever=base_retriever
 )
-# -----------------------------------------
 
 llm = ChatGoogleGenerativeAI(
     model=llm_model,
     google_api_key=GOOGLE_API_KEY,
-    temperature=0.0,
+    temperature=0.5,
 )
 
 store = {}