pavanjava · Dec 1, 2024 · Nov 26, 2024 · Nov 28, 2024 · Nov 28, 2024 · Nov 30, 2024
diff --git a/bootstraprag/cli.py b/bootstraprag/cli.py
@@ -38,6 +38,8 @@ def create(project_name, framework, template, observability):
             'rag-with-controllable-agents',
             'rag-with-llama-parse',
             'rag-with-adjacent-context',
+            'rag-with-sub-question-query-engine',
+            'rag-with-recursive-retriever',
             'rag-with-citation',
             'agents-with-introspection',
             'llama-deploy-with-simplemq',

diff --git a/bootstraprag/templates/llamaindex/rag_with_recursive_retriever/readme.md b/bootstraprag/templates/llamaindex/rag_with_recursive_retriever/readme.md
@@ -0,0 +1,17 @@
+### Recursive Retrieve Agents
+In LlamaIndex, the Recursive Retriever is a specialized component designed to enhance information retrieval by navigating through interconnected nodes within a document or across multiple documents. Unlike traditional retrieval methods that fetch information based solely on direct relevance, the Recursive Retriever delves deeper into the relationships between data points, allowing for a more comprehensive extraction of pertinent information.
+
+### How to run?
+`pip install -r requirements.txt`
+`python main.py`
+
+### How to expose as API?
+`python api_server.py`
+- Method: POST
+- API: http://localhost:8000/api/v1/chat-completion
+- Body:
+```json
+{
+  "query": "explain mlops architecture"
+}
+```
diff --git a/...prag/templates/llamaindex/rag_with_recursive_retriever/recursive_retriever_agents_core.py b/...prag/templates/llamaindex/rag_with_recursive_retriever/recursive_retriever_agents_core.py
@@ -27,6 +27,7 @@ def __init__(self, agent_names: List[str], data_dir: str = 'data'):
         self.document_data = {}
         self.agents = {}
         self.query_engine = None
+        self.client: qdrant_client.QdrantClient = None
 
         # Load environment variables
         load_dotenv(find_dotenv())
@@ -60,6 +61,7 @@ def _setup_vector_store(self):
             url=os.environ['DB_URL'],
             api_key=os.environ['DB_API_KEY']
         )
+        self.client = client
         self.vector_store = QdrantVectorStore(
             client=client,
             collection_name=os.environ['COLLECTION_NAME']
@@ -91,11 +93,16 @@ def _create_query_engine_tools(self, agent_name: str, vector_index: VectorStoreI
     def _build_agents(self):
         """Build agents with their respective tools."""
         for agent_name in self.agent_names:
-            # Build indices
-            vector_index = VectorStoreIndex.from_documents(
-                self.document_data[agent_name],
-                storage_context=self.storage_context
-            )
+
+            if not self.client.collection_exists(collection_name=os.environ.get("COLLECTION_NAME")):
+                # Build indices
+                vector_index = VectorStoreIndex.from_documents(
+                    self.document_data[agent_name],
+                    storage_context=self.storage_context
+                )
+            else:
+                vector_index = VectorStoreIndex.from_vector_store(vector_store=self.vector_store)
+
             summary_index = SummaryIndex.from_documents(
                 self.document_data[agent_name],
                 storage_context=self.storage_context

diff --git a/.../templates/llamaindex/rag_with_recursive_retriever_with_observability/readme.md b/.../templates/llamaindex/rag_with_recursive_retriever_with_observability/readme.md
@@ -0,0 +1,24 @@
+### Recursive Retrieve Agents
+In LlamaIndex, the Recursive Retriever is a specialized component designed to enhance information retrieval by navigating through interconnected nodes within a document or across multiple documents. Unlike traditional retrieval methods that fetch information based solely on direct relevance, the Recursive Retriever delves deeper into the relationships between data points, allowing for a more comprehensive extraction of pertinent information.
+
+### How to run?
+`pip install -r requirements.txt`
+`python main.py`
+
+### How to expose as API?
+`python api_server.py`
+- Method: POST
+- API: http://localhost:8000/api/v1/chat-completion
+- Body:
+```json
+{
+  "query": "explain mlops architecture"
+}
+```
+
+#### How to spin observability
+- run `docker compose -f docker-compose-langfuse.yml up`
+- launch langfuse in browser `http://localhost:3000`
+- click on `signup`
+- create `organization` & `project`
+- once done create your `public` and `private` api keys
diff --git a/...prag/templates/llamaindex/rag_with_self_correction_with_observability/readme.md b/...prag/templates/llamaindex/rag_with_self_correction_with_observability/readme.md
@@ -5,7 +5,7 @@
 - click on `signup`
 - create `organization` & `project`
 - once done create your `public` and `private` api keys
-- 
+
 #### Instructions to run the code
 - Navigate to the root of the project and run the below command
 - `pip install -r requirements.txt`

diff --git a/bootstraprag/templates/llamaindex/rag_with_sub_question_query_engine/.env b/bootstraprag/templates/llamaindex/rag_with_sub_question_query_engine/.env
@@ -0,0 +1,29 @@
+DB_URL='http://localhost:6333'
+DB_API_KEY='th3s3cr3tk3y'
+COLLECTION_NAME='SUB_QUESTION_COLLECTION'
+
+OPENAI_API_KEY='sk-proj-'
+OPENAI_EMBED_MODEL='gpt-4o'
+
+# use this incase you are prefering to experiment with local models.
+OLLAMA_BASE_URL='http://localhost:11434'
+OLLAMA_LLM_MODEL='llama3.1'
+OLLAMA_EMBED_MODEL='nomic-embed-text:latest'
+
+CHUNK_SIZE=128
+CHUNK_OVERLAP=20
+
+# logger can be controlled usiing env
+CRITICAL = 50
+FATAL = 50
+ERROR = 40
+WARNING = 30
+WARN = 30
+INFO = 20
+DEBUG = 10
+NOTSET = 0
+
+LIT_SERVER_PORT=8000
+LIT_SERVER_WORKERS_PER_DEVICE=4
+
+IS_EVALUATION_NEEDED=true
diff --git a/bootstraprag/templates/llamaindex/rag_with_sub_question_query_engine/api_server.py b/bootstraprag/templates/llamaindex/rag_with_sub_question_query_engine/api_server.py
@@ -0,0 +1,29 @@
+from dotenv import load_dotenv, find_dotenv
+from sub_question_query_engine import SubQuestionQueryEngineAgent
+import litserve as lit
+import os
+
+
+class SubQuestionQueryAPI(lit.LitAPI):
+    def __init__(self):
+        load_dotenv(find_dotenv())
+        self.engine = None
+
+    def setup(self, device):
+        self.engine = SubQuestionQueryEngineAgent()
+
+    def decode_request(self, request, **kwargs):
+        return request['query']
+
+    def predict(self, x, **kwargs):
+        return self.engine.query(x)
+
+    def encode_response(self, output, **kwargs):
+        return {'Agent': output}
+
+
+if __name__ == "__main__":
+    lit_api = SubQuestionQueryAPI()
+    server = lit.LitServer(lit_api=lit_api, api_path='/api/v1/chat-completion',
+                           workers_per_device=int(os.environ.get('LIT_SERVER_WORKERS_PER_DEVICE')))
+    server.run(port=os.environ.get('LIT_SERVER_PORT'))
diff --git a/bootstraprag/templates/llamaindex/rag_with_sub_question_query_engine/data/orthodontics.pdf b/bootstraprag/templates/llamaindex/rag_with_sub_question_query_engine/data/orthodontics.pdf
diff --git a/bootstraprag/templates/llamaindex/rag_with_sub_question_query_engine/main.py b/bootstraprag/templates/llamaindex/rag_with_sub_question_query_engine/main.py
@@ -0,0 +1,45 @@
+from sub_question_query_engine import SubQuestionQueryEngineAgent
+
+
+def print_welcome_message():
+    print("\n=== Orthodontics Query System ===")
+    print("Type your question and press Enter")
+    print("Type 'quit' to exit the program")
+    print("================================\n")
+
+
+if __name__ == "__main__":
+    # Initialize the engine
+    engine = SubQuestionQueryEngineAgent()
+
+    # Load and index documents
+    print("Initializing the system... Please wait...")
+    engine.load_and_index_documents()
+
+    # Display welcome message
+    print_welcome_message()
+
+    while True:
+        try:
+            # Get user input
+            user_question = input("\nEnter your question: ").strip()
+
+            # Check for quit command
+            if user_question.lower() == 'quit':
+                print("\nThank you for using the system. Goodbye!")
+                break
+
+            # Skip empty questions
+            if not user_question:
+                print("Please enter a valid question.")
+                continue
+
+            # Execute query and print response
+            print("\nProcessing your question...\n")
+            response = engine.query(user_question)
+            print("\nResponse:", response)
+            print("\n" + "-" * 50)  # Separator line
+
+        except Exception as e:
+            print(f"\nAn error occurred: {str(e)}")
+            print("Please try again with a different question.")
diff --git a/bootstraprag/templates/llamaindex/rag_with_sub_question_query_engine/readme.md b/bootstraprag/templates/llamaindex/rag_with_sub_question_query_engine/readme.md
@@ -0,0 +1,17 @@
+### Sub Question Query Engine
+The Sub-Question Query Engine in LlamaIndex is designed to handle complex queries that require information from multiple data sources. It operates by decomposing a complex query into several sub-questions, each directed to the most relevant data source. After obtaining responses to these sub-questions, it synthesizes them into a comprehensive final answer.
+
+### How to run?
+`pip install -r requirements.txt`
+`python main.py`
+
+### How to expose as API?
+`python api_server.py`
+- Method: POST
+- API: http://localhost:8000/api/v1/chat-completion
+- Body:
+```json
+{
+  "query": "Explain vertical plane in orthodontics"
+}
+```
diff --git a/bootstraprag/templates/llamaindex/rag_with_sub_question_query_engine/requirements.txt b/bootstraprag/templates/llamaindex/rag_with_sub_question_query_engine/requirements.txt
@@ -1 +1,6 @@
-llamaindex
+llamaindex
+llama-index-llms-ollama
+llama-index-llms-openai
+llama-index-agent-openai
+llama-index-vector-stores-qdrant
+qdrant-client
diff --git a/...prag/templates/llamaindex/rag_with_sub_question_query_engine/sub_question_query_engine.py b/...prag/templates/llamaindex/rag_with_sub_question_query_engine/sub_question_query_engine.py
@@ -0,0 +1,93 @@
+from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, StorageContext
+from llama_index.core.tools import QueryEngineTool, ToolMetadata
+from llama_index.core.query_engine import SubQuestionQueryEngine
+from llama_index.core.callbacks import CallbackManager, LlamaDebugHandler
+from llama_index.core import Settings
+from llama_index.embeddings.openai import OpenAIEmbedding
+from llama_index.llms.openai import OpenAI
+from llama_index.vector_stores.qdrant import QdrantVectorStore
+from dotenv import load_dotenv, find_dotenv
+import qdrant_client
+import os
+
+
+class SubQuestionQueryEngineAgent:
+    def __init__(self):
+        # Load environment variables
+        load_dotenv(find_dotenv())
+
+        # Initialize debug handler and settings
+        self._setup_settings()
+
+        # Initialize Qdrant clients
+        self.client = qdrant_client.QdrantClient(
+            url=os.environ['DB_URL'],
+            api_key=os.environ['DB_API_KEY']
+        )
+        self.aclient = qdrant_client.AsyncQdrantClient(
+            url=os.environ['DB_URL'],
+            api_key=os.environ['DB_API_KEY']
+        )
+
+        # Initialize query engine
+        self.query_engine = None
+
+    def _setup_settings(self):
+        """Configure LlamaIndex settings"""
+        llama_debug = LlamaDebugHandler(print_trace_on_end=True)
+        callback_manager = CallbackManager([llama_debug])
+
+        Settings.callback_manager = callback_manager
+        Settings.llm = OpenAI(
+            model=os.environ.get("OPENAI_EMBED_MODEL"),
+            temperature=0.0
+        )
+        Settings.embed_model = OpenAIEmbedding(model="text-embedding-3-small")
+        Settings.chunk_size = int(os.environ.get("CHUNK_SIZE"))
+        Settings.chunk_overlap = int(os.environ.get("CHUNK_OVERLAP"))
+
+    def load_and_index_documents(self, input_dir="data"):
+        """Load documents and create vector store index"""
+        # Load documents
+        orthodontics_docs = SimpleDirectoryReader(input_dir=input_dir).load_data(
+            show_progress=True
+        )
+
+        # Setup vector store and index
+        vector_store = QdrantVectorStore(
+            client=self.client,
+            aclient=self.aclient,
+            collection_name=os.environ.get("COLLECTION_NAME")
+        )
+        storage_context = StorageContext.from_defaults(vector_store=vector_store)
+        if not self.client.collection_exists(collection_name=os.environ.get("COLLECTION_NAME")):
+            vector_query_engine = VectorStoreIndex.from_documents(
+                documents=orthodontics_docs,
+                storage_context=storage_context
+            ).as_query_engine()
+        else:
+            vector_query_engine = VectorStoreIndex.from_vector_store(
+                vector_store=vector_store).as_query_engine()
+
+        # Setup query engine tools
+        query_engine_tools = [
+            QueryEngineTool(
+                query_engine=vector_query_engine,
+                metadata=ToolMetadata(
+                    name="orthodontics_tool",
+                    description="guide on the orthodontics",
+                ),
+            ),
+        ]
+
+        # Initialize SubQuestionQueryEngine
+        self.query_engine = SubQuestionQueryEngine.from_defaults(
+            query_engine_tools=query_engine_tools
+        )
+
+    def query(self, question: str):
+        """Execute a query and return the response"""
+        if self.query_engine is None:
+            raise ValueError("Query engine not initialized. Call load_and_index_documents first.")
+
+        return self.query_engine.query(question)
diff --git a/bootstraprag/templates/llamaindex/rag_with_sub_question_query_engine_with_observability/.env b/bootstraprag/templates/llamaindex/rag_with_sub_question_query_engine_with_observability/.env
@@ -0,0 +1,34 @@
+DB_URL='http://localhost:6333'
+DB_API_KEY='th3s3cr3tk3y'
+COLLECTION_NAME='SUB_QUESTION_COLLECTION'
+
+OPENAI_API_KEY='sk-proj-'
+OPENAI_EMBED_MODEL='gpt-4o'
+
+# use this incase you are prefering to experiment with local models.
+OLLAMA_BASE_URL='http://localhost:11434'
+OLLAMA_LLM_MODEL='llama3.1'
+OLLAMA_EMBED_MODEL='nomic-embed-text:latest'
+
+CHUNK_SIZE=128
+CHUNK_OVERLAP=20
+
+# Langfuse Observability Details
+LANGFUSE_PUBLIC_KEY='pk-lf-'
+LANGFUSE_SECRET_KEY='sk-lf-'
+LANGFUSE_HOST='http://localhost:3000'
+
+# logger can be controlled usiing env
+CRITICAL = 50
+FATAL = 50
+ERROR = 40
+WARNING = 30
+WARN = 30
+INFO = 20
+DEBUG = 10
+NOTSET = 0
+
+LIT_SERVER_PORT=8000
+LIT_SERVER_WORKERS_PER_DEVICE=4
+
+IS_EVALUATION_NEEDED=true
diff --git a/...ag/templates/llamaindex/rag_with_sub_question_query_engine_with_observability/__init__.py b/...ag/templates/llamaindex/rag_with_sub_question_query_engine_with_observability/__init__.py
diff --git a/.../templates/llamaindex/rag_with_sub_question_query_engine_with_observability/api_server.py b/.../templates/llamaindex/rag_with_sub_question_query_engine_with_observability/api_server.py
@@ -0,0 +1,29 @@
+from dotenv import load_dotenv, find_dotenv
+from sub_question_query_engine import SubQuestionQueryEngineAgent
+import litserve as lit
+import os
+
+
+class SubQuestionQueryAPI(lit.LitAPI):
+    def __init__(self):
+        load_dotenv(find_dotenv())
+        self.engine = None
+
+    def setup(self, device):
+        self.engine = SubQuestionQueryEngineAgent()
+
+    def decode_request(self, request, **kwargs):
+        return request['query']
+
+    def predict(self, x, **kwargs):
+        return self.engine.query(x)
+
+    def encode_response(self, output, **kwargs):
+        return {'Agent': output}
+
+
+if __name__ == "__main__":
+    lit_api = SubQuestionQueryAPI()
+    server = lit.LitServer(lit_api=lit_api, api_path='/api/v1/chat-completion',
+                           workers_per_device=int(os.environ.get('LIT_SERVER_WORKERS_PER_DEVICE')))
+    server.run(port=os.environ.get('LIT_SERVER_PORT'))
diff --git a/...es/llamaindex/rag_with_sub_question_query_engine_with_observability/data/orthodontics.pdf b/...es/llamaindex/rag_with_sub_question_query_engine_with_observability/data/orthodontics.pdf