LLM-RAG-HW-0001

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+chroma_db/chroma.sqlite3 filter=lfs diff=lfs merge=lfs -text

`requirements.txt

@@ -0,0 +1,6 @@
+gradio
+chromadb
+langchain
+langchain-community
+langchain-huggingface
+sentence-transformers

app.py

@@ -0,0 +1,25 @@
+import os
+import gradio as gr
+from rag import get_rag_response 
+
+# --- Gradio Chat Wrapper ---
+def chat_function(message, history):
+    """
+    Gradio passes the user's new 'message' and the chat 'history' automatically.
+    We hand the message to your existing RAG pipeline.
+    """
+    answer = get_rag_response(message) 
+    return answer
+
+# --- UI Definition ---
+demo = gr.ChatInterface(
+    fn=chat_function,
+    title="My Personal Knowledge Base",
+    description="Ask my AI questions based on my personal notes.",
+    theme=gr.themes.Soft(),
+    examples=["What are my notes on machine learning?", "Summarize my recent project plans."]
+)
+
+if __name__ == "__main__":
+    # Hugging Face Spaces requires the app to bind to 0.0.0.0
+    demo.launch(server_name="0.0.0.0", server_port=7860)

chroma_db/09989bea-8a40-4c1f-9c95-96378d5756e5/data_level0.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cea68868543f4bba4ab1015429b2ddb2958c34359559d94b39ad58acbabecda7
+size 167600

chroma_db/09989bea-8a40-4c1f-9c95-96378d5756e5/header.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0e81c3b22454233bc12d0762f06dcca48261a75231cf87c79b75e69a6c00150
+size 100

chroma_db/09989bea-8a40-4c1f-9c95-96378d5756e5/length.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:65dac0c40a9671394281201f420c4685aed98abd8727c0244a9acca5a92c2a98
+size 400

chroma_db/09989bea-8a40-4c1f-9c95-96378d5756e5/link_lists.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
+size 0

chroma_db/chroma.sqlite3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:107f763763598bb28422390e9dbfbbcc7062b7688a6e6ec68bb6d1436baedf0a
+size 229376

rag.py

@@ -0,0 +1,63 @@
+import os
+from langchain_community.vectorstores import Chroma
+from langchain_community.embeddings import HuggingFaceEmbeddings
+from langchain_huggingface import HuggingFaceEndpoint
+from langchain_core.prompts import ChatPromptTemplate
+
+# --- Configuration ---
+CHROMA_PATH = "./chroma_db"
+
+PROMPT_TEMPLATE = """
+You are a helpful, private AI assistant. 
+Answer the question based ONLY on the following context from my personal notes:
+
+{context}
+
+---
+
+Question: {question}
+"""
+
+# 1. Initialize DB and Embeddings ONCE globally
+print("Loading embeddings and database...")
+embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
+db = Chroma(persist_directory=CHROMA_PATH, embedding_function=embeddings)
+
+# 2. Connect to Hugging Face's Cloud LLM instead of local Ollama
+hf_token = os.environ.get("HF_TOKEN")
+if not hf_token:
+    raise ValueError("HF_TOKEN missing. Please add your Hugging Face Access Token to the Space Secrets.")
+
+llm = HuggingFaceEndpoint(
+    repo_id="google/gemma-2-2b-it",
+    task="text-generation",
+    max_new_tokens=512,
+    huggingfacehub_api_token=hf_token
+)
+
+def get_rag_response(query_text):
+    """This function is called by your Gradio app.py"""
+    
+    # Retrieve the relevant chunks
+    results = db.similarity_search_with_relevance_scores(query_text, k=3)
+
+    if len(results) == 0 or results[0][1] < 0.2:
+        return "I couldn't find any highly relevant notes to answer that."
+
+    # Format the chunks
+    context_text = "\n\n---\n\n".join([doc.page_content for doc, _ in results])
+
+    # Build the prompt
+    prompt_template = ChatPromptTemplate.from_template(PROMPT_TEMPLATE)
+    prompt = prompt_template.format(context=context_text, question=query_text)
+
+    # Ask Gemma
+    response_text = llm.invoke(prompt)
+
+    # Format the final output to return to the Gradio UI
+    final_answer = response_text.strip() + "\n\n### 📚 Sources Used:\n"
+    for doc, score in results:
+        source_name = doc.metadata.get('source', 'Unknown')
+        final_answer += f"- **{source_name}** (Relevance: {score:.2f})\n"
+
+    return final_answer