tree-convertion added

Json_2_tree.py

@@ -0,0 +1,22 @@
+
+# Function to recursively build the tree graph from the JSON
+def json_to_dot(graph, node_id, parent_node, parent_label):
+            if isinstance(parent_node, dict):
+                for key, value in parent_node.items():
+                    if key.startswith("Question"):
+                        question_id = f"{node_id}_{key}"
+                        label_text = "\n".join(value[i:i+30] for i in range(0, len(value), 30))
+                        shape = 'diamond' if len(value) > 50 else 'box'
+                        graph.node(question_id, label_text, shape=shape, style='filled', fillcolor='lightblue')
+                        graph.edge(parent_label, question_id, color='black')
+                        json_to_dot(graph, question_id, value, question_id)
+                    elif key in ["Yes", "No"]:
+                        option_label = f"{node_id}_{key}"
+                        graph.node(option_label, key, shape='box', style='filled', fillcolor='lightgreen' if key == "Yes" else 'lightcoral')
+                        graph.edge(parent_label, option_label, label=key, color='black')
+                        json_to_dot(graph, option_label, value, option_label)
+                    elif key == "Result":
+                        result_label = f"{node_id}_{key}"
+                        result_str = f"{key}: {value}\nCouncil regulations: {parent_node['Council regulations']}"
+                        graph.node(result_label, result_str, shape='box', style='filled', fillcolor='lightgrey')
+                        graph.edge(parent_label, result_label, color='black')

app.py

@@ -1,8 +1,154 @@
+from beyondllm import source,retrieve,embeddings,llms,generator
+import os
+from getpass import getpass
+from beyondllm.vectordb import ChromaVectorDb
+import json
+from graphviz import Digraph
+import graphviz
+import streamlit as st
+from beyondllm.llms import AzureOpenAIModel
+from beyondllm.embeddings import AzureAIEmbeddings
+from Scripts.Json_to_tree import create_decision_tree_image
+import networkx as nx
+import matplotlib.pyplot as plt
 
-from fastapi import FastAPI
+import os
+os.environ["PATH"] += os.pathsep + 'Graphiviz/Graphviz-11.0.0-win64/bin/bin/'
+os.environ["PATH"] += os.pathsep + "Graphviz2.38/bin/dot.exe"
 
-app = FastAPI()
+st.title("Chat with document")
 
-@app.get("/")
-def greet_json():
-    return {"Hello": "World!"}
+st.text("Enter API Key")
+
+# api_key = st.text_input("API Key:", type="password")
+# os.environ['OPENAI_API_KEY'] = api_key
+st.success("API Key entered successfully!")
+
+st.caption("Upload a PDF document to get information from the document.")
+uploaded_file = st.file_uploader("Choose a PDF file", type='pdf')
+submit=st.button("Get the data")
+if submit:
+
+    question = "Give Decision taken in the document"
+    system_prompt = '''You are a business analyst with extensive knowledge of legal documents and regulatory documentation.
+    Your expertise is in helping people understand and extract key information from such documents. 
+    Your task is to extract the rules and exceptions in a way that enables the creation of a decision tree, facilitating integration into the proper flow.
+    Legal Document Context: {context}
+    
+    Create a decision tree in JSON format based on the following structure:
+
+    Write a question and question should be two response like yes or no. if yes it has fallowing answers or other question 
+        - If Yes, the result should be: "Not restricted" additional -Council regulations: provide dates and articles if possible.
+        - If No, proceed to the next question2.( by giving some link to the next question not direct to next question)
+
+    2. Next question based on the previous question outcome.
+        - If Yes, the result should be: "Not restricted" additional -Council regulations: provide dates and articles if possible.
+        - If No, proceed to the next question.
+    In simple terms - flow chat if conditons.
+    [Continue this structure for as many questions as needed, ensuring each question branches into Yes/No answers and provides appropriate results based on the Council regulations.]
+    Please continue this format for as many questions as needed, ensuring each question follows the same structure.
+    Output is the JSON response follow this pattern: Do not change everytime Json output
+    This is JSON output Example, add more questions in this formate only. 
+        {
+        "Question1": ,
+        "Yes": {
+            "Result": ,
+            "Council regulations": 
+        },
+        "No": {
+            "Question2": ,
+            "Yes": {
+                "Result":,
+                "Council regulations": 
+            },
+            "No": {
+                "Question3": ,
+                "Yes": {
+                    "Result": ,
+                    "Council regulations":
+                },
+                "No": {
+                    "Result": ,
+                    "Council regulations": 
+                }
+            }
+        }
+    }
+    Additional Instructions:
+    
+    Analyze the entire document to identify all relevant rules and exceptions.
+    Ensure that the descriptions of rules and exceptions are clear and concise.
+    Include relevant dates, jurisdictions, and specific regulations where applicable.
+    Structure the questions and answers to facilitate the creation of a logical decision tree or workflow.
+    If the regulation mentions specific products, territories, or operations, include them in the appropriate sections.
+    Aim to simplify legal language while maintaining accuracy and intent.
+    [Provide your answer in JSON form. Reply with only the answer in JSON form and include no other commentary]:
+    Provide your answer in JSON form. Reply with only the answer in JSON form and include no other commentary
+
+    Return Valid Json to create Tree 
+    '''
+
+    
+
+    if uploaded_file is not None and question:
+
+        save_path = "./uploaded_files"
+        if not os.path.exists(save_path):
+            os.makedirs(save_path)
+        file_path = os.path.join(save_path, uploaded_file.name)
+        with open(file_path, "wb") as f:
+            f.write(uploaded_file.getbuffer())
+
+        data = source.fit(file_path, dtype="pdf", chunk_size=1024, chunk_overlap=0)
+        embed_model = AzureAIEmbeddings(
+                endpoint_url="https://marketplace.openai.azure.com/",
+                azure_key="d6d9522a01c74836907af2f3fd72ff85",
+                api_version="2024-02-01",
+                deployment_name="text-embed-marketplace")
+
+        retriever = retrieve.auto_retriever(data, embed_model, type="normal", top_k=4)
+        # vectordb = ChromaVectorDb(collection_name="my_persistent_collection", persist_directory="./db/chroma/")
+
+        # llm = llms.ChatOpenAIModel()
+        BASE_URL = "https://gpt-res.openai.azure.com/"
+        DEPLOYMENT_NAME= "gpt-4-32k" 
+        API_KEY = "a20bc67dbd7c47ed8c978bbcfdacf930"
+        llm = AzureOpenAIModel(model="gpt4",azure_key = API_KEY,deployment_name=DEPLOYMENT_NAME ,endpoint_url=BASE_URL,model_kwargs={"max_tokens":512,"temperature":0.1})
+        pipeline = generator.Generate(question=question, system_prompt=system_prompt, retriever=retriever, llm=llm)
+        decision_tree_json = pipeline.call()
+        response = json.loads(decision_tree_json)
+        # Function to recursively create DOT format from JSON
+        def json_to_dot(graph, node_id, parent_node, parent_label):
+            if isinstance(parent_node, dict):
+                for key, value in parent_node.items():
+                    if key.startswith("Question"):
+                        question_id = f"{node_id}_{key}"
+                        label_text = "\n".join(value[i:i+30] for i in range(0, len(value), 30))
+                        shape = 'diamond' if len(value) > 50 else 'box'
+                        graph.node(question_id, label_text, shape=shape, style='filled', fillcolor='lightblue')
+                        graph.edge(parent_label, question_id, color='black')
+                        json_to_dot(graph, question_id, value, question_id)
+                    elif key in ["Yes", "No"]:
+                        option_label = f"{node_id}_{key}"
+                        graph.node(option_label, key, shape='box', style='filled', fillcolor='lightgreen' if key == "Yes" else 'lightcoral')
+                        graph.edge(parent_label, option_label, label=key, color='black')
+                        json_to_dot(graph, option_label, value, option_label)
+                    elif key == "Result":
+                        result_label = f"{node_id}_{key}"
+                        result_str = f"{key}: {value}\nCouncil regulations: {parent_node['Council regulations']}"
+                        graph.node(result_label, result_str, shape='box', style='filled', fillcolor='lightgrey')
+                        graph.edge(parent_label, result_label, color='black')
+
+        # Create a new graph
+        dot = graphviz.Digraph(comment='Decision Tree')
+        # Add the root node
+        dot.node('Root', 'Start', shape='ellipse', style='filled', fillcolor='lightyellow')
+        # Build the DOT format
+        json_to_dot(dot, "Root", response, "Root")
+        # Render and display the graph using Graphviz engine
+        dot.format = 'png'
+        dot.render('decision_tree', view=True)
+        import streamlit as st
+        with st.chat_message(""):
+            st.write("")
+            st.image('decision_tree.png', caption='tree from json')

requirements.txt

@@ -1,3 +1,7 @@
 
-fastapi
-uvicorn[standard]
+beyondllm
+streamlit
+graphviz
+llama_index.vector_stores.chroma
+llama-index-embeddings-huggingface   
+llama-index-embeddings-azure_openai