[INIT]

.gitignore

@@ -0,0 +1,2 @@
+vector_store/db/chroma.sqlite3
+vector_store/db/a83981c6-dd1a-4d14-9eba-0296b4da1fd2/data_level0.bin

app.py

@@ -0,0 +1,96 @@
+
+# # Cell 2 - Login Setup
+# from huggingface_hub import login
+# from dotenv import load_dotenv
+# import os
+# load_dotenv()
+
+# login(token=os.getenv("HUGGINGFACEHUB_API_TOKEN"), add_to_git_credential=True)
+
+
+# from embedding import embeddings
+# from db.chroma import load_and_setup_db,search_cases
+# from chat.chat.hermes_llm import ChatManager
+
+# VECTOR_DB_PATH = os.getenv("VECTOR_DB_PATH")
+
+# vector_store = load_and_setup_db(VECTOR_DB_PATH,embeddings)
+
+
+# query = "somthing"
+
+# result = search_cases(vectorstore=vector_store,query=query,k=1)
+# legal_chat = ChatManager(temperature=0.1)
+
+# respose = legal_chat.get_response(legal_chat[0]['content'],query=query)
+
+import gradio as gr
+import os
+from huggingface_hub import login
+from dotenv import load_dotenv
+from embedding import embeddings
+from db.chroma import load_and_setup_db, search_cases
+from chat.hermes_llm import ChatManager
+
+# Load environment variables
+load_dotenv()
+
+# Login to Hugging Face
+login(token=os.getenv("HUGGINGFACEHUB_API_TOKEN"), add_to_git_credential=True)
+
+# Initialize components
+VECTOR_DB_PATH = os.getenv("VECTOR_DB_PATH")
+vector_store = load_and_setup_db(VECTOR_DB_PATH, embeddings)
+legal_chat = ChatManager(temperature=0.1)
+
+def process_query(query, chat_history):
+    try:
+        # Search relevant cases
+        results = search_cases(vectorstore=vector_store, query=query, k=1)
+        response=None
+        if len(results)>0:
+        # Get response from chat manager
+            response = legal_chat.get_response(results[0]['content'], query=query)
+        else :
+            response = "No Document match"
+        # Update chat history
+        chat_history.append((query, response))
+        return "", chat_history
+    except Exception as e:
+        return "", chat_history + [(query, f"Error: {str(e)}")]
+
+# Create Gradio interface
+with gr.Blocks(title="Legal Chat Assistant") as demo:
+    gr.Markdown("# Legal Chat Assistant")
+    gr.Markdown("Ask questions about legal cases and get AI-powered responses.")
+    
+    chatbot = gr.Chatbot(
+        [],
+        elem_id="chatbot",
+        bubble_full_width=False,
+        height=400
+    )
+    
+    with gr.Row():
+        query_input = gr.Textbox(
+            placeholder="Enter your query here...",
+            show_label=False,
+            scale=4
+        )
+        submit_btn = gr.Button("Send", scale=1)
+    
+ 
+    # Set up event handlers
+    submit_btn.click(
+        process_query,
+        inputs=[query_input, chatbot],
+        outputs=[query_input, chatbot]
+    )
+    query_input.submit(
+        process_query,
+        inputs=[query_input, chatbot],
+        outputs=[query_input, chatbot]
+    )
+
+if __name__ == "__main__":
+    demo.launch(share=True)

chat/hermes_llm.py

@@ -0,0 +1,49 @@
+
+from langchain_community.chat_models import ChatOpenAI
+from langchain_core.prompts import ChatPromptTemplate
+from typing import List, Dict
+import os
+OPENROUTER_API_KEY = os.getenv("OPENROUTER_API_KEY")
+OPENROUTER_API_BASE = "https://openrouter.ai/api/v1"
+DEFAULT_MODEL = "nousresearch/hermes-3-llama-3.1-405b:free"
+
+class ChatManager:
+    def __init__(self, temperature: float = 0.7):
+        self.chat = ChatOpenAI(
+            openai_api_base=OPENROUTER_API_BASE,
+            openai_api_key=OPENROUTER_API_KEY,
+            model_name=DEFAULT_MODEL,
+            temperature=temperature,
+        )
+        
+        self.system_message = """You are an advanced assistant designed to help users by retrieving the most relevant information from a predefined set of documents or cases and then providing an accurate response based on that data.
+
+Your job is as follows:
+
+1. When the user submits a query, match the query with the most relevant case from the database.
+2. Extract only the denoised and contextually relevant text from that case.
+3. Use that extracted text to answer the user's query with precision and clarity. 
+4. If the relevant text isn't found, let the user know that the information is not available or ask for more clarification.
+5. Avoid providing information outside the scope of the retrieved text.
+
+Always focus on relevance and clarity in your response, maintaining coherence with the user's original query."""
+
+        self.user_message = """
+Context: {context}
+        
+Query: {query}
+"""
+        
+        self.prompt = ChatPromptTemplate.from_messages([
+            ("system", self.system_message),
+            ("human", self.user_message),
+        ])
+
+    def get_response(self, context: str, query: str) -> str:
+        prompt_value = self.prompt.invoke({
+            "context": context,
+            "query": query
+        })
+        messages = prompt_value.to_messages()
+        response = self.chat(messages)
+        return response.content

db/chroma.py

@@ -0,0 +1,109 @@
+from langchain.vectorstores import Chroma
+from langchain.embeddings import HuggingFaceEmbeddings
+from typing import List, Dict, Optional
+from tqdm import tqdm
+
+def load_and_setup_db(
+    persist_directory: str,
+    embeddings
+) -> Chroma:
+    """
+    Load the previously created ChromaDB with the same embedding model.
+    
+    Args:
+        persist_directory: Directory where the database is stored
+        embedding_model_name: Name of the embedding model to use
+        
+    Returns:
+        Chroma: Loaded vector store
+    """
+    
+    # Load the existing database
+    vectorstore = Chroma(
+        embedding_function=embeddings,
+        persist_directory=persist_directory
+    )
+    
+    return vectorstore
+
+def search_cases(
+    vectorstore: Chroma,
+    query: str,
+    k: int = 5,
+    metadata_filter: Optional[Dict] = None,
+    score_threshold: Optional[float] = 0.0
+) -> List[Dict]:
+    """
+    Search the database for relevant cases.
+    
+    Args:
+        vectorstore: Loaded Chroma vector store
+        query: Search query text
+        k: Number of results to return
+        metadata_filter: Optional filter for metadata fields
+        score_threshold: Minimum similarity score threshold
+        
+    Returns:
+        List of relevant documents with scores and metadata
+    """
+    # Perform similarity search with metadata filtering
+    docs_and_scores = vectorstore.similarity_search_with_score(
+        query,
+        k=k,
+        filter=metadata_filter
+    )
+    
+    # Process and filter results
+    results = []
+    for doc, score in docs_and_scores:
+        # Convert score to similarity (assuming distance score)
+        similarity = 1 - score
+        
+        # Apply score threshold
+        if score_threshold and similarity < score_threshold:
+            continue
+            
+        result = {
+            'content': doc.page_content,
+            'metadata': doc.metadata,
+            'similarity_score': round(similarity, 4)
+        }
+        results.append(result)
+    if len(results)==0 and len(docs_and_scores)>0:
+        results.append(docs_and_scores[0])
+    return results
+
+# Example usage function
+def search_and_display_results(
+    vectorstore: Chroma,
+    query: str,
+    k: int = 5,
+    metadata_filter: Optional[Dict] = None,
+    score_threshold: float = 0.7
+) -> None:
+    """
+    Search and display results in a formatted way.
+    """
+    print(f"\nSearching for: {query}")
+    print("-" * 50)
+    
+    results = search_cases(
+        vectorstore=vectorstore,
+        query=query,
+        k=k,
+        metadata_filter=metadata_filter,
+        score_threshold=score_threshold
+    )
+    
+    if not results:
+        print("No matching results found.")
+        return
+        
+    print(f"Found {len(results)} relevant matches:\n")
+    
+    for i, result in enumerate(results, 1):
+        print(f"Match {i}:")
+        print(f"Similarity Score: {result['similarity_score']}")
+        print(f"Metadata: {result['metadata']}")
+        print(f"Content: {result['content'][:200]}...")  # Show first 200 chars
+        print("-" * 50)

embedding.py

@@ -0,0 +1,8 @@
+from langchain.embeddings import HuggingFaceInferenceAPIEmbeddings
+from dotenv import load_dotenv
+import os
+load_dotenv()
+
+embeddings = HuggingFaceInferenceAPIEmbeddings(
+    api_key=os.getenv("HUGGINGFACEHUB_API_TOKEN"), model_name="sentence-transformers/all-MiniLM-L6-v2"
+)

requirements.txt

@@ -0,0 +1,4 @@
+langchain 
+langchain_community 
+chromadb 
+python-dotenv 

vector_store/db/a83981c6-dd1a-4d14-9eba-0296b4da1fd2/header.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83384ec1ec80f415f9b29c84d9ef966930faf31d936cfa2c22eb0b0c9bf59924
+size 100

vector_store/db/a83981c6-dd1a-4d14-9eba-0296b4da1fd2/index_metadata.pickle

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e29cdd64643b3a13902e63a70336a1287517f59e4873145c60c0f1179e68b795
+size 752123

vector_store/db/a83981c6-dd1a-4d14-9eba-0296b4da1fd2/length.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24281a3e8212857faa150d8b5f534cf0847f3971bd40a1144149b57c10410bbf
+size 52000

vector_store/db/a83981c6-dd1a-4d14-9eba-0296b4da1fd2/link_lists.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:667a7d6a2ad2c3ab267035a264717476a2da859dd8cde2d1565c2bfb65610908
+size 110344