ADDED: query db endpoint

README.md

@@ -189,6 +189,16 @@ percentage, missing keywords, and profile summary.
   - Returns articles and research findings related to the specified tech topic.
   - Handles exceptions gracefully and returns error messages in JSON format.
 
+### 22. Query Database
+
+- **Route:** `/query_db`
+- **Description:** This API endpoint facilitates querying SQL databases using a Cohere ReAct Agent, integrated with Langchain's SQLDBToolkit.
+- **Feature:**
+  - Upload a `.db` file containing the database to be queried.
+  - Provide a natural language prompt or query to retrieve relevant information
+  from the database.
+  - Utilizes a Cohere ReAct Agent to process user queries and generate responses.
+
 ## Usage
 
 Each endpoint accepts specific parameters as described in the respective endpoint documentation. Users can make POST requests to these endpoints with the required parameters to perform the desired tasks.
@@ -204,7 +214,7 @@ Each endpoint accepts specific parameters as described in the respective endpoin
 1. Clone the repository:
 
    ```bash
-   git clone https://github.com/AquibPy/LLM-use-cases-API.git
+   git clone https://github.com/AquibPy/Genify.git
    ```
 
 2. Install dependencies:
@@ -213,15 +223,15 @@ Each endpoint accepts specific parameters as described in the respective endpoin
     pip install -r requirements.txt
     ```
 
-3. Create ```.evn``` file
+3. Create ```.env``` file
 
     Save Google and Hugging Face API key in this file.
 
 ## Running the Server
 
 Start the FastAPI server by running the following command:
     ```
-    uvicorn main:app --reload
+    fastapi run api.py
     ```
 
 ## Contributing

api.py

@@ -15,9 +15,9 @@
 from mongo import MongoDB
 from helper_functions import get_qa_chain,get_gemini_response,get_url_doc_qa,extract_transcript_details,\
     get_gemini_response_health,get_gemini_pdf,read_sql_query,remove_substrings,questions_generator,groq_pdf,\
-    summarize_audio,chatbot_send_message,extraxt_pdf_text,advance_rag_llama_index
+    summarize_audio,chatbot_send_message,extraxt_pdf_text,advance_rag_llama_index,parse_sql_response
 from langchain_groq import ChatGroq
-from langchain.chains import ConversationChain
+from langchain.chains.conversation.base import ConversationChain
 from langchain.chains.conversation.memory import ConversationBufferWindowMemory
 from langchain_core.prompts import ChatPromptTemplate
 from auth import create_access_token
@@ -30,6 +30,14 @@
 from sendgrid.helpers.mail import Mail
 from uuid import uuid4
 # from tech_news_agent.crew import run_crew
+from langchain.agents import AgentExecutor
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_cohere.react_multi_hop.agent import create_cohere_react_agent
+from langchain_cohere.chat_models import ChatCohere
+from langchain_community.utilities.sql_database import SQLDatabase
+from langchain_community.agent_toolkits import SQLDatabaseToolkit
+import tempfile
+import shutil
 
 
 os.environ["LANGCHAIN_TRACING_V2"]="true"
@@ -654,6 +662,61 @@ async def get_data(endpoint_name: str, token: str = Depends(oauth2_scheme)):
 #         return ResponseText(response=output)
 #     except Exception as e:
 #         return {"error": str(e)}
+
+@app.post("/query_db",description="""
+          The Query Database endpoint provides a service for interacting with SQL databases using a Cohere ReAct Agent. 
+          It leverages Langchain's existing SQLDBToolkit to answer questions and perform queries over SQL database.
+          """)
+async def query_db(database: UploadFile = File(...), prompt: str = Form(...)):
+    try: 
+        with tempfile.NamedTemporaryFile(delete=False, suffix='.' + database.filename.split('.')[-1]) as temp_file:
+            shutil.copyfileobj(database.file, temp_file)
+            db_path = temp_file.name
+
+        llm = ChatCohere(model="command-r-plus", temperature=0.1, verbose=True,cohere_api_key=os.getenv("COHERE_API_KEY"))
+        db = SQLDatabase.from_uri(f"sqlite:///{db_path}")
+        toolkit = SQLDatabaseToolkit(db=db, llm=llm)
+        context = toolkit.get_context()
+        tools = toolkit.get_tools()
+        chat_prompt = ChatPromptTemplate.from_template("{input}")
+
+        agent = create_cohere_react_agent(
+            llm=llm,
+            tools=tools,
+            prompt=chat_prompt
+        )
+        agent_executor = AgentExecutor(
+            agent=agent,
+            tools=tools,
+            verbose=True,
+            return_intermediate_steps=False,
+        )
+
+        preamble = settings.QUERY_DB_PROMPT.format(schema_info=context)
+
+        out = agent_executor.invoke({
+           "input": prompt,
+           "preamble": preamble
+        })
+
+        output  = parse_sql_response(out["output"])
+        db = MongoDB()
+        payload = {
+            "endpoint": "/query_db",
+            "input": prompt,
+            "output": output
+        }
+        mongo_data = {"Document": payload}
+        result = db.insert_data(mongo_data)
+        print(result)
+
+        return ResponseText(response=output)
+
+    except Exception as e:
+        raise Exception(f"Error handling uploaded file: {e}")
+
+    finally:
+        database.file.close()
 
 if __name__ == '__main__':
     import uvicorn

helper_functions.py

@@ -317,7 +317,26 @@ def advance_rag_llama_index(pdf,model,question):
     os.remove(tmp_path)
 
     return str(response)
-
+import re
+def parse_sql_response(response):
+    # Split the response into individual SQL statements
+    sql_statements = re.split(r"(?<=\*\/)\n\n+", response)
+
+    # Format each SQL statement
+    formatted_sql_statements = []
+    for sql_statement in sql_statements:
+        if sql_statement.strip():
+            # Remove comments
+            sql_statement = re.sub(r'/\*.*?\*/', '', sql_statement, flags=re.DOTALL)
+            # Replace newlines and tabs with spaces
+            sql_statement = sql_statement.replace('\n', ' ').replace('\t', ' ')
+            # Add a newline after each semicolon
+            sql_statement = re.sub(r';(?!\s*CREATE|INSERT|SELECT|UPDATE|DELETE)', ';\n', sql_statement)
+            formatted_sql_statements.append(sql_statement.strip())
+
+    # Join the formatted SQL statements into a single string
+    formatted_response = '\n'.join(formatted_sql_statements)
+    return formatted_response
 
 if __name__ == "__main__":
     create_vector_db()

requirements.txt

@@ -27,4 +27,5 @@ llama-index-embeddings-langchain
 grpcio
 # crewai
 # crewai-tools
-proto-plus
+proto-plus
+langchain-cohere

settings.py

@@ -101,4 +101,22 @@
 navigating issues of climate change, political polarization, and the ethical implications of artificial intelligence.
 As we reflect on the journey of humanity, from ancient civilizations to the digital age, we are reminded of our shared past and
 the collective responsibility to shape a more equitable and sustainable future.
-"""
+"""
+
+QUERY_DB_PROMPT = """## Task And Context
+        You use your advanced complex reasoning capabilities to help people by answering their questions and other requests interactively. 
+        You will be asked a very wide array of requests on all kinds of topics. You will be equipped with a wide range of search engines or similar tools to help you,
+        which you use to research your answer. You may need to use multiple tools in parallel or sequentially to complete your task. 
+        You should focus on serving the user's needs as best you can, which will be wide-ranging.
+
+        ## Style Guide
+        Unless the user asks for a different style of answer, you should answer in full sentences, using proper grammar and spelling.
+
+        ## Additional Information
+        You are an expert who answers the user's question by creating SQL queries and executing them.
+        You are equipped with a number of relevant SQL tools.
+        You should also present the SQL query used to provide the data.
+
+        Here is information about the database:
+        {schema_info}
+        """