+# MAGIC
+# MAGIC
+# MAGIC
+# MAGIC
+# MAGIC -- Don't forget the tracker
+# MAGIC 
+
+# COMMAND ----------
+
+#Note: don't pin databricks langchain/agent/mlflow, pin the rest
+%pip install -U --quiet databricks-sdk==0.40.0 databricks-langchain databricks-agents mlflow[databricks] databricks-vectorsearch==0.49 langchain==0.3.19 langchain_core==0.3.37 bs4==0.0.2 markdownify==0.14.1 pydantic==2.10.1
+dbutils.library.restartPython()
+
+# COMMAND ----------
+
+#Note: Load the dataset directly as delta table
+%run ../_resources/00-init $reset_all_data=false
+
+# COMMAND ----------
+
+# MAGIC %sql
+# MAGIC -- Note: Run some select to review the different dataset available
+# MAGIC SELECT * FROM databricks_documentation ...
+
+# COMMAND ----------
+
+# MAGIC %md
+# MAGIC ## Creating our tools: Using Unity Catalog Functions
+# MAGIC
+# MAGIC Let's start by defining the functions our LLM will be able to execute. These functions can contain any logic, from simple SQL to advanced python.
+
+# COMMAND ----------
+
+# MAGIC %md
+# MAGIC ### First tool: Computing xxx
+# MAGIC
+# MAGIC xxx what are we doing
+# MAGIC
+# MAGIC We'll save this function within Unity Catalog. You can open the explorer to review the functions created in this notebook.
+# MAGIC
+# MAGIC *Note: This is a very simple first example for this demo. We'll implement a broader math tool later on.*
+
+# COMMAND ----------
+
+# MAGIC %sql
+# MAGIC CREATE OR REPLACE FUNCTION query_customer_details(size_in_inch FLOAT COMMENT 'xxxx')
+# MAGIC RETURNS FLOAT
+# MAGIC LANGUAGE SQL
+# MAGIC COMMENT 'convert size from inch to cm'
+# MAGIC RETURN xxx
+# MAGIC
+# MAGIC -- let's test our function:
+# MAGIC SELECT convert_inch_to_cm(10) as 10_inches_in_cm;
+
+# COMMAND ----------
+
+# MAGIC %md
+# MAGIC ### Second tool: Computing xxx
+# MAGIC
+# MAGIC xxx what are we doing
+# MAGIC
+# MAGIC We'll save this function within Unity Catalog. You can open the explorer to review the functions created in this notebook.
+# MAGIC
+# MAGIC *Note: This is a very simple first example for this demo. We'll implement a broader math tool later on.*
+
+# COMMAND ----------
+
+# MAGIC %sql
+# MAGIC CREATE OR REPLACE FUNCTION query_customer_invoice_history(size_in_inch FLOAT COMMENT 'xxxx')
+# MAGIC RETURNS FLOAT
+# MAGIC LANGUAGE SQL
+# MAGIC COMMENT 'convert size from inch to cm'
+# MAGIC RETURN xxx
+# MAGIC
+# MAGIC -- let's test our function:
+# MAGIC SELECT convert_inch_to_cm(10) as 10_inches_in_cm;
+
+# COMMAND ----------
+
+# MAGIC %md
+# MAGIC ### Last tool: Generate coupon xxx
+# MAGIC
+# MAGIC xxx what are we doing
+# MAGIC
+# MAGIC We'll save this function within Unity Catalog. You can open the explorer to review the functions created in this notebook.
+# MAGIC
+# MAGIC *Note: This is a very simple first example for this demo. We'll implement a broader math tool later on.*
+
+# COMMAND ----------
+
+TODO mockup python function to show how to have a tool in pyton
+
+%sql
+CREATE OR REPLACE FUNCTION compute_math(expr STRING)
+RETURNS STRING
+LANGUAGE PYTHON
+COMMENT 'Run any mathematical function and returns the result as output. Supports python syntax like math.sqrt(13)'
+AS
+$$
+ import ast
+ import operator
+ import math
+ try:
+ if expr.startswith('```') and expr.endswith('```'):
+ expr = expr[3:-3].strip()
+ node = ast.parse(expr, mode='eval').body
+ return eval_node(node)
+ except Exception as ex:
+ return str(ex)
+$$;
+-- let's test our function:
+SELECT compute_math("(2+2)/3") as result;
+
+# COMMAND ----------
+
+# MAGIC %md
+# MAGIC ### ... Continue, add the tool to UC,
+
+# COMMAND ----------
+
+# MAGIC %md
+# MAGIC ### ... Try on Playground
+
+# COMMAND ----------
+
+# MAGIC %md
+# MAGIC ### ... Try on Playground
+
+# COMMAND ----------
+
+from langchain.agents import AgentExecutor, create_tool_calling_agent
+prompt = get_prompt(prompt="You are an assistant for a python developer. Internally, you have a tool named execute_python_code that can generate and run python code to help answering what the customer is asking. input: valid python code as a string. output: the result of the return value from the code execution. Don't mention you have tools or the tools name. Make sure you send the full python code at once to the function and that the code has a return statement at the end to capture the result. Don't print anything in the code you write, return the result you need as final instruction. Make sure the python code is valid. Only send python. Here is an example: 'def square_function(number):\\n return number*number\\n\\nreturn square_function(3)'")
+tools = get_tools()
+agent = create_tool_calling_agent(llm, tools, prompt)
+agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)
+
+#Let's print the answer! Note that the LLM often makes a few error, but then analyze it and self-correct.
+answer = agent_executor.invoke({"input": "What's the result of the fibonacci suite? Display its result for 5."})
+displayHTML(answer['output'].replace('\n', '+ Ask me anything and I'll do my best to help you. +
+{{ content }}
+
+
+
+ + Ask a question to see the AI reasoning process +
+'+(a?i:Pn(i,!0))+`"+(a?i:Pn(i,!0))+`+${this.parser.parse(t)}+`}html({text:t}){return t}heading({tokens:t,depth:n}){return`
${this.parser.parseInline(t)}
+`}table(t){let n="",a="";for(let i=0;i${Pn(t,!0)}`}br(t){return"An error occurred:
"+Pn(a.message+"",!0)+"";return n?Promise.resolve(l):l}if(n)return Promise.reject(a);throw a}}}const Da=new mS;function He(e,t){return Da.parse(e,t)}He.options=He.setOptions=function(e){return Da.setOptions(e),He.defaults=Da.defaults,Vm(He.defaults),He};He.getDefaults=Su;He.defaults=Wa;He.use=function(...e){return Da.use(...e),He.defaults=Da.defaults,Vm(He.defaults),He};He.walkTokens=function(e,t){return Da.walkTokens(e,t)};He.parseInline=Da.parseInline;He.Parser=rn;He.parser=rn.parse;He.Renderer=po;He.TextRenderer=Tu;He.Lexer=sn;He.lexer=sn.lex;He.Tokenizer=bo;He.Hooks=Jl;He.parse=He;He.options;He.setOptions;He.use;He.walkTokens;He.parseInline;rn.parse;sn.lex;Y(null);const dr={listeners:new Map,addListener(e,t){this.listeners.set(e,t)},removeListener(e){this.listeners.delete(e)},emit(e,t){const n=this.listeners.get(e);n&&n(t)}},hS=async(e,t,n=!0,a="telco")=>{var l;try{const i=await fetch("/api/agent/chat",{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({messages:e,intelligence_enabled:n,use_case:a})});if(!i.ok)throw new Error("Network response was not ok");const o=(l=i.body)==null?void 0:l.getReader();if(!o)throw new Error("No reader available");const s=new TextDecoder;let r="";for(;;){const{done:u,value:c}=await o.read();if(u)break;r+=s.decode(c,{stream:!0});const d=r.split(` +`);r=d.pop()||"";for(const v of d)if(v.startsWith("data: "))try{const m=v.slice(6),h=JSON.parse(m);dr.emit(t,h)}catch(m){console.error("Error parsing SSE data:",m)}}}catch(i){throw console.error("Error in sendMessageToAgent:",i),i}},Rd=async(e="telco")=>{try{const t=await fetch(`/api/agent/questions?use_case=${e}`);if(!t.ok)throw new Error("Failed to fetch questions");return await t.json()}catch(t){return console.error("Error fetching predefined questions:",t),[]}},gS={xmlns:"http://www.w3.org/2000/svg",width:"1em",height:"1em",viewBox:"0 0 24 24",class:"icon-brain-circuit"},yS=Yn({__name:"IconBrainCircuit",setup(e){return(t,n)=>(Be(),Ze("svg",gS,n[0]||(n[0]=[ke("g",{fill:"none",stroke:"currentColor","stroke-linecap":"round","stroke-linejoin":"round","stroke-width":"2"},[ke("path",{d:"M12 4.5a2.5 2.5 0 0 0-4.96-.462A2.5 2.5 0 0 0 4.5 8.5a2.5 2.5 0 0 0 1.17 4.896c.401.665 1.208 1.056 2.047 1.102C7.745 16.7 8.874 19 12 19a2.5 2.5 0 0 0 4.96.462A2.5 2.5 0 0 0 19.5 15.5a2.5 2.5 0 0 0-1.17-4.896c-.401-.665-1.208-1.056-2.047-1.102C16.255 7.3 15.126 5 12 5"}),ke("path",{d:"M12 7v7m-3-3h6m-8-4h2m12 0h-2m-1 8h-2m-10 0h2"})],-1)])))}});const bS=Vi(yS,[["__scopeId","data-v-a229375b"]]),pS={class:"message-content"},kS={class:"message-avatar"},SS={class:"text-white"},xS={class:"message-bubble"},wS={key:0,class:"message-with-intelligence"},CS=["innerHTML"],_S=["onClick"],VS=["innerHTML"],PS={class:"message-time"},IS={key:0,class:"message bot-message typing-animation"},TS={class:"message-content"},AS={class:"message-avatar"},ES={class:"message-bubble"},$S=["innerHTML"],BS={class:"message-time"},DS={key:1,class:"message bot-message typing-indicator"},RS={class:"message-content"},MS={class:"message-avatar"},LS={class:"chat-input"},OS={class:"d-flex align-center"},FS={class:"predefined-questions mt-3"},NS={class:"d-flex flex-wrap"},Md=2,Ld=4,zS=Yn({__name:"ChatBot",props:{thinking:{type:Boolean},agents:{},finalAnswer:{},finalInformations:{},intelligenceEnabled:{type:Boolean}},emits:["update:thinking","update:agents","update:final-answer","update:final-informations","use-case-change"],setup(e,{expose:t,emit:n}){const a=e,l=n,i=Y(new Map),o=Y([]),s=Y(""),r=Y([]),u=()=>Date.now().toString()+Math.random().toString(36).substring(2,9),c=Y([{id:u(),text:"Hello! I'm your AI assistant. How can I help you today?",sender:"bot",timestamp:new Date}]),d=Y(""),v=Y(null),m=Y(!1),h=Y(!1),g=Y(""),y=Y(!1),p=Y(""),k=/(\S+\s*)/g,x=Y(null),P=Y(""),I=Y([]),A=Y("telco"),$=async z=>{A.value=z,c.value=[{id:u(),text:"Hello! I'm your AI assistant. How can I help you today?",sender:"bot",timestamp:new Date}],o.value=[],s.value="",r.value=[],m.value=!1,h.value=!1,l("update:thinking",!1),l("update:agents",[]),l("update:final-answer",""),l("update:final-informations",[]),i.value.clear();try{const M=await Rd(z);I.value=M}catch(M){console.error("Error fetching predefined questions:",M),I.value=[]}await Ee(),v.value&&(v.value.scrollTop=0)};yt(async()=>{I.value=await Rd(A.value),O()});const C=z=>{d.value=z.text,Ee(()=>{E()})},w=z=>z.map(M=>({role:M.sender==="user"?"user":"assistant",content:M.text})),E=async()=>{if(!d.value.trim())return;const z=u();c.value.push({id:z,text:d.value,sender:"user",timestamp:new Date}),d.value,d.value="",await O(),m.value=!0,h.value=!0,l("update:thinking",!0),l("update:agents",[]);try{o.value=[],s.value="",r.value=[];const M=w(c.value);dr.addListener(z,V),await hS(M,z,a.intelligenceEnabled,A.value),dr.removeListener(z)}catch(M){console.error("Error getting bot response:",M),m.value=!1,h.value=!1,l("update:thinking",!1),c.value.push({id:u(),text:"I'm sorry, I encountered an error. Please try again.",sender:"bot",timestamp:new Date}),await O()}},V=z=>{var M;switch(z.type){case"thinking-start":o.value=[],l("update:agents",[]),l("update:final-answer",""),l("update:final-informations",[]),m.value=!0;break;case"tool":a.intelligenceEnabled&&(o.value=[...o.value,z.data],l("update:agents",o.value));break;case"final-answer":P.value=u(),s.value=z.data.final_answer,r.value=z.data.final_informations||[],l("update:final-answer",s.value),l("update:final-informations",r.value),i.value.set(P.value,{question:((M=c.value[c.value.length-1])==null?void 0:M.text)||"",tools:o.value,final_answer:s.value,final_informations:r.value,non_intelligent_answer:z.data.final_answer}),_(z.data.final_answer,!a.intelligenceEnabled),m.value=!1;break}},_=async(z,M=!1)=>{const F=P.value;c.value.push({id:F,text:"",sender:"bot",timestamp:new Date}),y.value=!1,p.value=z,g.value="",m.value=!1,await O();const q=z.match(k)||[],re=M?Ld*2:Ld,ue=M?Md/2:Md,be=c.value.findIndex(K=>K.id===F);if(be!==-1){for(let K=0;K
+
+# COMMAND ----------
+
+# MAGIC %run ../config
+
+# COMMAND ----------
+
+from pyspark.sql.functions import pandas_udf
+import pandas as pd
+import pyspark.sql.functions as F
+from pyspark.sql.functions import col, udf, length, pandas_udf
+import os
+import mlflow
+import yaml
+from typing import Iterator
+from mlflow import MlflowClient
+mlflow.set_registry_uri('databricks-uc')
+
+# Workaround for a bug fix that is in progress
+mlflow.spark.autolog(disable=True)
+
+import warnings
+warnings.filterwarnings("ignore")
+# Disable MLflow warnings
+import logging
+logging.getLogger('mlflow').setLevel(logging.ERROR)
+
+# COMMAND ----------
+
+def use_and_create_db(catalog, dbName, cloud_storage_path = None):
+ print(f"USE CATALOG `{catalog}`")
+ spark.sql(f"USE CATALOG `{catalog}`")
+ spark.sql(f"""create database if not exists `{dbName}` """)
+
+assert catalog not in ['hive_metastore', 'spark_catalog'], "Please use a UC schema"
+#If the catalog is defined, we force it to the given value and throw exception if not.
+if len(catalog) > 0:
+ current_catalog = spark.sql("select current_catalog()").collect()[0]['current_catalog()']
+ if current_catalog != catalog:
+ catalogs = [r['catalog'] for r in spark.sql("SHOW CATALOGS").collect()]
+ if catalog not in catalogs:
+ spark.sql(f"CREATE CATALOG IF NOT EXISTS {catalog}")
+ use_and_create_db(catalog, dbName)
+
+
+print(f"using catalog.database `{catalog}`.`{dbName}`")
+spark.sql(f"""USE `{catalog}`.`{dbName}`""")
+
+# COMMAND ----------
+
+# DBTITLE 1,Optional: Allowing Model Serving IPs
+#If your workspace has ip access list, you need to allow your model serving endpoint to hit your AI gateway. Based on your region, IPs might change. Please reach out your Databrics Account team for more details.
+
+#def allow_serverless_ip():
+# from databricks.sdk import WorkspaceClient
+# from databricks.sdk.service import settings
+#
+# w = WorkspaceClient()
+#
+# # cleanup
+# w.ip_access_lists.delete(ip_access_list_id='xxxx')
+# created = w.ip_access_lists.create(label=f'serverless-model-serving',
+# ip_addresses=['xxxx/32'],
+# list_type=settings.ListType.ALLOW)
+
+# COMMAND ----------
+
+# MAGIC %md
+# MAGIC
+# MAGIC ## Helpers to get catalog and index status:
+
+# COMMAND ----------
+
+# Helper function
+def get_latest_model_version(model_name):
+ mlflow_client = MlflowClient(registry_uri="databricks-uc")
+ latest_version = 1
+ for mv in mlflow_client.search_model_versions(f"name='{model_name}'"):
+ version_int = int(mv.version)
+ if version_int > latest_version:
+ latest_version = version_int
+ return latest_version
+
+# COMMAND ----------
+
+# DBTITLE 1,endpoint
+import time
+
+def endpoint_exists(vsc, vs_endpoint_name):
+ try:
+ return vs_endpoint_name in [e['name'] for e in vsc.list_endpoints().get('endpoints', [])]
+ except Exception as e:
+ #Temp fix for potential REQUEST_LIMIT_EXCEEDED issue
+ if "REQUEST_LIMIT_EXCEEDED" in str(e):
+ print("WARN: couldn't get endpoint status due to REQUEST_LIMIT_EXCEEDED error. The demo will consider it exists")
+ return True
+ else:
+ raise e
+
+def wait_for_vs_endpoint_to_be_ready(vsc, vs_endpoint_name):
+ for i in range(180):
+ try:
+ endpoint = vsc.get_endpoint(vs_endpoint_name)
+ except Exception as e:
+ #Temp fix for potential REQUEST_LIMIT_EXCEEDED issue
+ if "REQUEST_LIMIT_EXCEEDED" in str(e):
+ print("WARN: couldn't get endpoint status due to REQUEST_LIMIT_EXCEEDED error. Please manually check your endpoint status")
+ return
+ else:
+ raise e
+ status = endpoint.get("endpoint_status", endpoint.get("status"))["state"].upper()
+ if "ONLINE" in status:
+ return endpoint
+ elif "PROVISIONING" in status or i <6:
+ if i % 20 == 0:
+ print(f"Waiting for endpoint to be ready, this can take a few min... {endpoint}")
+ time.sleep(10)
+ else:
+ raise Exception(f'''Error with the endpoint {vs_endpoint_name}. - this shouldn't happen: {endpoint}.\n Please delete it and re-run the previous cell: vsc.delete_endpoint("{vs_endpoint_name}")''')
+ raise Exception(f"Timeout, your endpoint isn't ready yet: {vsc.get_endpoint(vs_endpoint_name)}")
+
+# COMMAND ----------
+
+# DBTITLE 1,index
+def index_exists(vsc, endpoint_name, index_full_name):
+ try:
+ vsc.get_index(endpoint_name, index_full_name).describe()
+ return True
+ except Exception as e:
+ if 'RESOURCE_DOES_NOT_EXIST' not in str(e):
+ print(f'Unexpected error describing the index. This could be a permission issue.')
+ raise e
+ return False
+
+def wait_for_index_to_be_ready(vsc, vs_endpoint_name, index_name):
+ for i in range(180):
+ idx = vsc.get_index(vs_endpoint_name, index_name).describe()
+ index_status = idx.get('status', idx.get('index_status', {}))
+ status = index_status.get('detailed_state', index_status.get('status', 'UNKNOWN')).upper()
+ url = index_status.get('index_url', index_status.get('url', 'UNKNOWN'))
+ if "ONLINE" in status:
+ return
+ if "UNKNOWN" in status:
+ print(f"Can't get the status - will assume index is ready {idx} - url: {url}")
+ return
+ elif "PROVISIONING" in status:
+ if i % 40 == 0: print(f"Waiting for index to be ready, this can take a few min... {index_status} - pipeline url:{url}")
+ time.sleep(10)
+ else:
+ raise Exception(f'''Error with the index - this shouldn't happen. DLT pipeline might have been killed.\n Please delete it and re-run the previous cell: vsc.delete_index("{index_name}, {vs_endpoint_name}") \nIndex details: {idx}''')
+ raise Exception(f"Timeout, your index isn't ready yet: {vsc.get_index(index_name, vs_endpoint_name)}")
+
+def wait_for_model_serving_endpoint_to_be_ready(ep_name):
+ from databricks.sdk import WorkspaceClient
+ from databricks.sdk.service.serving import EndpointStateReady, EndpointStateConfigUpdate
+ import time
+
+ # TODO make the endpoint name as a param
+ # Wait for it to be ready
+ w = WorkspaceClient()
+ state = ""
+ for i in range(200):
+ state = w.serving_endpoints.get(ep_name).state
+ if state.config_update == EndpointStateConfigUpdate.IN_PROGRESS:
+ if i % 40 == 0:
+ print(f"Waiting for endpoint to deploy {ep_name}. Current state: {state}")
+ time.sleep(10)
+ elif state.ready == EndpointStateReady.READY:
+ print('endpoint ready.')
+ return
+ else:
+ break
+ raise Exception(f"Couldn't start the endpoint, timeout, please check your endpoint for more details: {state}")
+
+# COMMAND ----------
+
+#Display a better quota message
+def display_quota_error(e, ep_name):
+ if "QUOTA_EXCEEDED" in str(e):
+ displayHTML(f'Please select another endpoint in the ../config file (VECTOR_SEARCH_ENDPOINT_NAME="
+
+# COMMAND ----------
+
+VECTOR_SEARCH_ENDPOINT_NAME="dbdemos_vs_endpoint"
+
+catalog = "main__build"
+dbName = db = "dbdemos_ai_agent"
+
+# COMMAND ----------
+
+# MAGIC %md
+# MAGIC ### License
+# MAGIC This demo installs the following external libraries on top of DBR(ML):
+# MAGIC
+# MAGIC
+# MAGIC | Library | License |
+# MAGIC |---------|---------|
+# MAGIC | langchain | [MIT](https://github.com/langchain-ai/langchain/blob/master/LICENSE) |
+# MAGIC | lxml | [BSD-3](https://pypi.org/project/lxml/) |
+# MAGIC | transformers | [Apache 2.0](https://github.com/huggingface/transformers/blob/main/LICENSE) |
+# MAGIC | unstructured | [Apache 2.0](https://github.com/Unstructured-IO/unstructured/blob/main/LICENSE.md) |
+# MAGIC | llama-index | [MIT](https://github.com/run-llama/llama_index/blob/main/LICENSE) |
+# MAGIC | tesseract | [Apache 2.0](https://github.com/tesseract-ocr/tesseract/blob/main/LICENSE) |
+# MAGIC | poppler-utils | [MIT](https://github.com/skmetaly/poppler-utils/blob/master/LICENSE) |
+# MAGIC | textstat | [MIT](https://pypi.org/project/textstat/) |
+# MAGIC | tiktoken | [MIT](https://github.com/openai/tiktoken/blob/main/LICENSE) |
+# MAGIC | evaluate | [Apache2](https://pypi.org/project/evaluate/) |
+# MAGIC | torch | [BDS-3](https://github.com/intel/torch/blob/master/LICENSE.md) |
+# MAGIC | tiktoken | [MIT](https://github.com/openai/tiktoken/blob/main/LICENSE) |
+# MAGIC
+# MAGIC
+# MAGIC
+# MAGIC