diff --git a/code/app/api.py b/code/app/api.py new file mode 100644 index 0000000..5c52571 --- /dev/null +++ b/code/app/api.py @@ -0,0 +1,35 @@ +from fastapi import FastAPI +from pydantic import BaseModel +from .backend import BDAgent + +app = FastAPI() + + +@app.get("/") +def health(): + return {"status": "ok"} + + +class QueryRequest(BaseModel): + query: str + history: str = "" + model: str = "anthropic.claude-haiku-4-5-20251001-v1:0" + base: str = "bacio_transacional_loja_app" + + +class QueryResponse(BaseModel): + response: str + input_tokens: int + output_tokens: int + total_tokens: int + + +@app.post("/agent", response_model=QueryResponse) +def run_agent(request: QueryRequest): + result = BDAgent.main(request.query, request.history, request.model, request.base) + return QueryResponse( + response=result["response"], + input_tokens=result["input_tokens"], + output_tokens=result["output_tokens"], + total_tokens=result["total_tokens"], + ) diff --git a/code/app/backend/BDAgent.py b/code/app/backend/BDAgent.py index 8d2e367..bb1ee2c 100644 --- a/code/app/backend/BDAgent.py +++ b/code/app/backend/BDAgent.py @@ -4,7 +4,6 @@ LangGraph Agent using AWS Bedrock Cross-Region Inference Profile with Tools This script demonstrates how to create a LangGraph agent that uses an AWS Bedrock inference profile with custom tools (add and multiply). """ - import boto3 from typing import TypedDict, Annotated from langgraph.graph import StateGraph, END @@ -18,9 +17,109 @@ from langfuse import Langfuse from langfuse.langchain import CallbackHandler from botocore.exceptions import ClientError import os -from backend.utils import dynamodb_read_table as drt WORKGROUP = "iceberg-workgroup" DATABASE = "dnx_warehouse" +TABLE = "poc_dnx_monthly_summary" +REGION = "us-east-1" + +# DynamoDB client +dynamodb = boto3.resource("dynamodb", region_name=REGION) + + +@tool +def get_monthly_report(id: str, variable: str) -> str: + """ + Get a specific variable's data from DynamoDB for a specific id. + + Args: + id: The id of the data + variable: The variable/column name to retrieve from the table + + Returns: + The content of the specified variable for the given id + """ + print(f"\n🔧 [TOOL CALLED] get_monthly_report for month: {id}, variable: {variable}") + + try: + table = dynamodb.Table(TABLE) + response = table.get_item(Key={"id": id}) + + if "Item" not in response: + return f"No report found for month: {id}" + + item = response["Item"] + content = item.get(variable, "") + + if not content: + return f"Variable '{variable}' not found for month: {id}" + + result = f"<{id}>\n{content}\n" + return result + + except ClientError as e: + error_message = e.response["Error"]["Message"] + return f"Error fetching report: {error_message}" + + +@tool +def get_consolidated_keys(id: str) -> str: + """ + Get the list of consolidated keys (variables) available in the table for a specific month. + + Args: + id: The id of the data + + Returns: + The list of available variables/keys for the specified data + """ + print(f"\n🔧 [TOOL CALLED] get_consolidated_keys for id: {id}") + + try: + table = dynamodb.Table(TABLE) + response = table.get_item(Key={"id": id}) + + if "Item" not in response: + return f"No data found for month: {id}" + + item = response["Item"] + chaves_consolidadas = item.get("chaves_consolidadas", "") + + if not chaves_consolidadas: + return f"No consolidated keys found for id: {id}" + + return chaves_consolidadas + + except ClientError as e: + error_message = e.response["Error"]["Message"] + return f"Error fetching consolidated keys: {error_message}" + + +def get_contexto() -> dict: + """ + Get contexto, filter, and items_disponiveis from DynamoDB where id=DASHBOARD+'_contexto'. + + Returns: + Dict with 'contexto', 'filter', and 'items_disponiveis' keys + """ + try: + table = dynamodb.Table(TABLE) + response = table.get_item(Key={"id": DASHBOARD + "_contexto"}) + + if "Item" not in response: + return {"contexto": "", "filter": "", "items_disponiveis": {}} + + item = response["Item"] + return { + "contexto": item.get("contexto", ""), + "filter": item.get("filter_key", ""), + "items_disponiveis": item.get("itens_disponiveis", {}), + } + + except ClientError as e: + error_message = e.response["Error"]["Message"] + return {"contexto": f"Error: {error_message}", "filter": "", "items_disponiveis": {}} + + def get_secret(): secret_name = "assistente-db-secrets-manager" @@ -112,7 +211,7 @@ def create_bedrock_llm(model_id: str, region: str = "us-east-1"): """ # Determine provider and model_kwargs based on model ID MODEL_ARNS = { - "anthropic.claude-haiku-4-5-20251001-v1:0": "arn:aws:bedrock:us-east-1:305427701314:inference-profile/global.anthropic.claude-haiku-4-5-20251001-v1:0", + "anthropic.claude-haiku-4-5-20251001-v1:0": "arn:aws:bedrock:us-east-1:305427701314:inference-profile/us.anthropic.claude-haiku-4-5-20251001-v1:0", "anthropic.claude-sonnet-4-5-20250929-v1:0": "arn:aws:bedrock:us-east-1:305427701314:inference-profile/global.anthropic.claude-sonnet-4-5-20250929-v1:0", "meta.llama4-maverick-17b-instruct-v1:0": "arn:aws:bedrock:us-east-1:305427701314:inference-profile/us.meta.llama4-maverick-17b-instruct-v1:0", "meta.llama4-scout-17b-instruct-v1:0": "arn:aws:bedrock:us-east-1:305427701314:inference-profile/us.meta.llama4-scout-17b-instruct-v1:0", @@ -130,7 +229,7 @@ def create_bedrock_llm(model_id: str, region: str = "us-east-1"): "amazon.nova-2-lite-v1:0": "amazon" } prefix={ - "anthropic.claude-haiku-4-5-20251001-v1:0": "global", + "anthropic.claude-haiku-4-5-20251001-v1:0": "us", "anthropic.claude-sonnet-4-5-20250929-v1:0": "global", "meta.llama4-maverick-17b-instruct-v1:0": "us", "meta.llama4-scout-17b-instruct-v1:0": "us", @@ -147,8 +246,7 @@ def create_bedrock_llm(model_id: str, region: str = "us-east-1"): ) # Bind tools to the LLM - #tools = [consult_answers,count_table_rows] - tools=[] + tools = [get_monthly_report, get_consolidated_keys] llm_with_tools = llm.bind_tools(tools) return llm_with_tools @@ -160,9 +258,7 @@ def call_model(state: AgentState, llm) -> AgentState: print(f"[MODEL] Calling Bedrock inference profile...") messages = state["messages"] - langfuse_handler = CallbackHandler() - config = {"configurable": {"thread_id": "abc123"},"callbacks": [langfuse_handler]} - response = llm.invoke(messages,config=config) + response = llm.invoke(messages) state["current_step"] = "model_called" return {"messages": [response]} @@ -180,6 +276,8 @@ def call_tools(state: AgentState) -> AgentState: tool_messages = [] tools_map = { + "get_monthly_report": get_monthly_report, + "get_consolidated_keys": get_consolidated_keys } # Execute each tool call @@ -267,24 +365,52 @@ def create_agent(inference_profile_arn: str, region: str = "us-east-1"): return app -def main(user_query,history,model): +def main(user_query,history,model,base): """Main execution function.""" + global DASHBOARD + DASHBOARD = base # Configuration - Update with your actual inference profile ARN - + INFERENCE_PROFILE_ARN = model REGION = "us-east-1" # System prompt for the agent - SYSTEM_PROMPT=""" You are a analitical agent, with acess to monthly reports about Bacio di latte + contexto_data = get_contexto() + if contexto_data["filter"]=="period": + CONSULT_RULES="""To use the tools you must give the id of the correspondant data, which can be associated to a given month and year in the following format year_month, which: + -Year is the year in 4 digits (2025,2024,2023,2022,2021,...) + -Month is th two digit representation: 01,02,03,04,05,06,07,08,09,10,11,12 + The format of the dict is: {id1:year_month1,id2:year_month2...} + Choose the correct id based on the following dict: +""" + elif contexto_data["filter"]=="event": + CONSULT_RULES="""To use the tools you must give the id of the correspondant data, which can be associated to a event, which is in the format "Name - City DD/MM/YYYY", where the last is a date in the format day/month/year. Theformat of elements in dict is {id1:event_description1,id2:event_description2...}""" + else: + CONSULT_RULES="""Wrong filter value, you must terminate the workflow and ask the user to contact the technical team""" + SYSTEM_PROMPT=""" You are a analitical agent in Brazilian Portuguese, with acess to monthly reports about a specific company, specified in the context. You have access to tools that lets you consult present variables in table, you always have access to "context", which keeps inside answers to different questions, that you may consult as you desire. + Do not access other variables besides the ones reported by the tool and "context". + You currently have access to data in a period specified in the context, so only answer questions inside the time window. - A Bacio di Latte é uma rede de gelaterias artesanais fundada em São Paulo, Brasil, em 2011, pelos irmãos milaneses Edoardo e Luigi Tonolli, que trouxeram a tradição do gelato italiano com ingredientes de alta qualidade, resultando em um sorvete cremoso e fresco, produzido diariamente, sem gordura hidrogenada ou trans, e que se tornou popular não só no Brasil, mas também nos EUA, representando uma experiência autêntica de gelato. - <\context> - - """+drt.read_table_as_xml("poc_dnx_monthly_summary","us-east-1")+"""" - <\reports> + """+contexto_data["contexto"]+""" + + """+CONSULT_RULES+""" + + """+str(contexto_data["items_disponiveis"])+""" + Here is the chat history:"""+history+""" - Aswer the user the best you can with the given information, if you don't know the answer or how to answer say so, only answer from what you know.""" + Inside the "NPS" in data is some useful values to calculate the NPS, which includes "distribuicao". + Inside of it are grades and the amount of people who given that grade. + Grades from 0 to 6 are detractors. + Grades from 7 to 8 are neutral. + Grades from 9 to 10 are promoters. + Calculate the percentage of them when prompted about NPS and then calculate the nps using the following formula: NPS = %promoter - %detractor, never use the medium of the notes. + + You have access to the tools: + -get_consolidated_keys: Given a id returns the column names inside of a entity of a given table element. + - get_monthly_report: given a id and a variable name, either one listed in the previous tool output or "context", returns its value. Using "context" gives you a summarization of many answers of questions asked to the customers. + Answer, in Brazilian Portuguese, to the user the best you can with the given information, if you don't know the answer or how to answer say so, only answer from what you know. + Always consult the most recent information when a date is not given, like questions "Quanto é meu nps?" """ print("=" * 60) print("LangGraph Agent with AWS Bedrock Inference Profile + Tools") @@ -297,11 +423,10 @@ def main(user_query,history,model): print("\nSystem Prompt: Configured ✓") print("=" * 60) - # Create the agent + # Create the agent with a unique session_id to group all steps + langfuse_handler = CallbackHandler() agent = create_agent(INFERENCE_PROFILE_ARN, REGION) - # Example query that requires tools - # Initialize state with system prompt initial_state = { "messages": [ @@ -314,8 +439,9 @@ def main(user_query,history,model): print(f"\nUser Query: {user_query}\n") print("-" * 60) - # Run the agent - final_state = agent.invoke(initial_state) + # Run the agent with callbacks at graph level + config = {"callbacks": [langfuse_handler], "tags": [DASHBOARD]} + final_state = agent.invoke(initial_state, config=config) # Display results print("-" * 60) @@ -336,7 +462,21 @@ def main(user_query,history,model): print("\n" + "=" * 60) print(f"Agent completed successfully. Final step: {final_state['current_step']}") + + # Aggregate token usage from all AIMessage objects + total_input_tokens = 0 + total_output_tokens = 0 + for msg in final_state["messages"]: + if isinstance(msg, AIMessage) and hasattr(msg, 'usage_metadata') and msg.usage_metadata: + total_input_tokens += msg.usage_metadata.get("input_tokens", 0) + total_output_tokens += msg.usage_metadata.get("output_tokens", 0) + langfuse.flush() - return final_state['messages'][-1].content + return { + "response": final_state['messages'][-1].content, + "input_tokens": total_input_tokens, + "output_tokens": total_output_tokens, + "total_tokens": total_input_tokens + total_output_tokens, + } if __name__=="__main__": - main("oi","ancar_nps_tradicional","","") \ No newline at end of file + main("Liste o nps mês a mês desde maio 2025 até dezembro 2025","","anthropic.claude-sonnet-4-5-20250929-v1:0") \ No newline at end of file diff --git a/code/app/front.py b/code/app/front.py index 71b2fa5..cc5b178 100644 --- a/code/app/front.py +++ b/code/app/front.py @@ -2,6 +2,7 @@ import streamlit as st import time from backend import BDAgent import boto3 +from boto3.dynamodb.conditions import Key # Configure the page - MUST BE FIRST st.set_page_config( @@ -11,9 +12,19 @@ st.set_page_config( ) session = boto3.Session() +dynamodb = boto3.resource("dynamodb", region_name="us-east-1") + +def list_bases(): + table = dynamodb.Table("poc_dnx_monthly_summary") + response = table.query( + IndexName="item_type_index", + KeyConditionExpression=Key("item_type").eq("contexto") + ) + return [item["id"].removesuffix("_contexto") for item in response.get("Items", [])] # Model selection with session state persistence MODELS = [ + "anthropic.claude-haiku-4-5-20251001-v1:0", "anthropic.claude-sonnet-4-5-20250929-v1:0", "meta.llama4-maverick-17b-instruct-v1:0", "meta.llama4-scout-17b-instruct-v1:0", @@ -27,7 +38,13 @@ selected_value = st.selectbox( MODELS, key="selected_model" ) - +BASES = list_bases() +BASES.sort() +base = st.selectbox( + "Selecione a base:", + BASES, + key="selected_base" +) # Initialize chat history in session state @@ -57,17 +74,19 @@ if prompt := st.chat_input("Type your message here..."): # Simulate streaming response (replace with actual API call) - full_response = BDAgent.main(prompt,str(st.session_state.messages),selected_value) - + result = BDAgent.main(prompt,str(st.session_state.messages),selected_value,base) + full_response = result["response"] + # Simulate typing effect displayed_response = "" for char in full_response: displayed_response += char message_placeholder.markdown(displayed_response + "▌") time.sleep(0.01) - + message_placeholder.markdown(full_response) - + st.caption(f"Tokens: {result['input_tokens']} in / {result['output_tokens']} out / {result['total_tokens']} total") + # Add assistant response to chat history st.session_state.messages.append({"role": "assistant", "content": full_response}) diff --git a/code/entrypoint.sh b/code/entrypoint.sh index f67db42..8f30279 100644 --- a/code/entrypoint.sh +++ b/code/entrypoint.sh @@ -1,23 +1,9 @@ #!/bin/bash -# Define o diretório base da aplicação dentro do contêiner -APP_DIR="/app" +# Inicia a API FastAPI em segundo plano +echo "Iniciando API FastAPI na porta 8000..." +uvicorn app.api:app --host 0.0.0.0 --port 8000 & -# Navega para o diretório do backend e inicia a API FastAPI em segundo plano -#echo "Iniciando API FastAPI na porta 8000..." -#python app/backend/main.py & - -# Aguarda alguns segundos para garantir que a API tenha tempo de iniciar -# Isso é opcional, mas pode ajudar a evitar problemas de conexão imediata do frontend -#echo "Aguardando a API iniciar..." -#sleep 10 # Ajuste o tempo conforme necessário - -# Navega para o diretório do frontend e inicia o aplicativo Streamlit em primeiro plano +# Inicia o aplicativo Streamlit em primeiro plano echo "Iniciando aplicativo Streamlit na porta 8501..." -# --server.headless=true é importante para rodar Streamlit em ambientes sem GUI (como Docker) -# --server.address=0.0.0.0 permite que o Streamlit seja acessado de fora do contêiner -# --server.enableCORS=false pode ser necessário dependendo da configuração, mas geralmente não para localhost streamlit run app/front.py --server.port 8501 --server.address 0.0.0.0 --server.headless true - -# O comando 'streamlit run' manterá o contêiner em execução. -# Se o Streamlit parar por algum motivo, o script e, consequentemente, o contêiner, terminarão. \ No newline at end of file diff --git a/code/requirements.txt b/code/requirements.txt index 36fb5ea..08de927 100644 --- a/code/requirements.txt +++ b/code/requirements.txt @@ -4,3 +4,5 @@ langgraph>=0.0.20 langchain>=0.1.0 streamlit langfuse +fastapi +uvicorn diff --git a/code/app/backend/utils/dynamodb_read_table.py b/code/utils/dynamodb_read_table.py similarity index 100% rename from code/app/backend/utils/dynamodb_read_table.py rename to code/utils/dynamodb_read_table.py diff --git a/infra/ecs_alb/Pulumi.Inovyo.yaml b/infra/ecs_alb/Pulumi.Inovyo.yaml index e301e79..af090dc 100644 --- a/infra/ecs_alb/Pulumi.Inovyo.yaml +++ b/infra/ecs_alb/Pulumi.Inovyo.yaml @@ -21,12 +21,12 @@ config: - 3.14.44.224/32 ecs_subnet_ids: - subnet-0f50f25a2fbb054d4 - - subnet-0014ea77951bbeb6c + - subnet-043a427630309c2f4 app-ecs:ecs: - task_name: assisnte-analitico-db-dev ecr_repo_name: assistente-analitico-db-dev ecr_image_tag: latest - ecr_image_digest: sha256:7a2aede33d8dd34822b73291d64e1ccce26980fc531290c54d992bcb7dee26fa + ecr_image_digest: sha256:c9cd83a7caf51e7eee471d47649bbe0157d4c2703f365010f12997e63d941d3d cpu: 256 memory: 512 desired_count: 1 @@ -36,11 +36,15 @@ config: min_capacity: 1 max_capacity: 3 target_value: 60.0 - lb_config: - name: listener + lb_configs: + - name: streamlit listener_port: 8501 target_port: 8501 container_port: 8501 + - name: api + listener_port: 8000 + target_port: 8000 + container_port: 8000 env_variables: LANGFUSE_HOST: http://172.31.252.176:3000 # SECRET_NAME: dev/ai-pge-doc-classification diff --git a/infra/ecs_alb/ecs.py b/infra/ecs_alb/ecs.py index 16f7169..3b70245 100644 --- a/infra/ecs_alb/ecs.py +++ b/infra/ecs_alb/ecs.py @@ -6,33 +6,42 @@ import ecr import json def deploy_app(config_ecs_app, app_ecs_cluster, alb_security_group, app_load_balancer_arn): - lb_config = config_ecs_app["lb_config"] - target_group = aws.lb.TargetGroup(f"app-target-group-{lb_config['listener_port']}", - port=lb_config["target_port"], - protocol="HTTP", - vpc_id=config.network["vpc_id"], - target_type="ip", - health_check=aws.lb.TargetGroupHealthCheckArgs( - path="/", # TODO if it doesn't work, can use /docs for fastapi - protocol="HTTP", - port="traffic-port", - healthy_threshold=2, - unhealthy_threshold=2, - timeout=5, - interval=30, - matcher="200-499", - ), - ) - aws.lb.Listener(f"app-listener-{lb_config['listener_port']}", - load_balancer_arn=app_load_balancer_arn, - port=lb_config["listener_port"], - protocol="HTTP", - default_actions=[aws.lb.ListenerDefaultActionArgs( - type="forward", - target_group_arn=target_group.arn, - )], - ) - # target_groups.append(target_group) + lb_configs = config_ecs_app["lb_configs"] + target_groups = [] + load_balancers = [] + + for lb_config in lb_configs: + tg = aws.lb.TargetGroup(f"app-target-group-{lb_config['listener_port']}", + port=lb_config["target_port"], + protocol="HTTP", + vpc_id=config.network["vpc_id"], + target_type="ip", + health_check=aws.lb.TargetGroupHealthCheckArgs( + path="/", + protocol="HTTP", + port="traffic-port", + healthy_threshold=2, + unhealthy_threshold=2, + timeout=5, + interval=30, + matcher="200-499", + ), + ) + aws.lb.Listener(f"app-listener-{lb_config['listener_port']}", + load_balancer_arn=app_load_balancer_arn, + port=lb_config["listener_port"], + protocol="HTTP", + default_actions=[aws.lb.ListenerDefaultActionArgs( + type="forward", + target_group_arn=tg.arn, + )], + ) + target_groups.append(tg) + load_balancers.append(aws.ecs.ServiceLoadBalancerArgs( + target_group_arn=tg.arn, + container_name=f"{config.project_name}-{config_ecs_app['task_name']}-{config.environment}-service", + container_port=lb_config["target_port"], + )) # Build and Push ECR # ecr_repos = ecr.create_ecr_repo(config_ecs_app['ecr_repo_name']) @@ -79,12 +88,11 @@ def deploy_app(config_ecs_app, app_ecs_cluster, alb_security_group, app_load_bal "cpu": 0, "portMappings": [ { - "name": "api", - "containerPort": lb_config["container_port"], - "hostPort": lb_config["target_port"], + "name": lb_cfg["name"], + "containerPort": lb_cfg["container_port"], + "hostPort": lb_cfg["target_port"], "protocol": "tcp", - } - # } for lb_config_item in lb_configs + } for lb_cfg in lb_configs ], "essential": True, "logConfiguration": { @@ -135,12 +143,6 @@ def deploy_app(config_ecs_app, app_ecs_cluster, alb_security_group, app_load_bal ) # ECS Service Setup - load_balancer = aws.ecs.ServiceLoadBalancerArgs( - target_group_arn=target_group.arn, - container_name=f"{config.project_name}-{config_ecs_app['task_name']}-{config.environment}-service", - container_port=lb_config["target_port"], - ) - app_service = aws.ecs.Service(f"{config.project_name}-{config_ecs_app['task_name']}-service", cluster=app_ecs_cluster.arn, task_definition=app_task_definition.arn, @@ -151,7 +153,7 @@ def deploy_app(config_ecs_app, app_ecs_cluster, alb_security_group, app_load_bal security_groups=[app_ecs_security_group.id], subnets=config.network["ecs_subnet_ids"], ), - load_balancers=[load_balancer], + load_balancers=load_balancers, scheduling_strategy="REPLICA", service_connect_configuration=aws.ecs.ServiceServiceConnectConfigurationArgs( enabled=True, diff --git a/infra/ecs_alb/iam.py b/infra/ecs_alb/iam.py index 59f5086..2226528 100644 --- a/infra/ecs_alb/iam.py +++ b/infra/ecs_alb/iam.py @@ -186,7 +186,10 @@ def create_task_role(): "dynamodb:Query", "dynamodb:DescribeTable" ], - "Resource": "arn:aws:dynamodb:us-east-1:305427701314:table/poc_dnx_monthly_summary" + "Resource": [ + "arn:aws:dynamodb:us-east-1:305427701314:table/poc_dnx_monthly_summary", + "arn:aws:dynamodb:us-east-1:305427701314:table/poc_dnx_monthly_summary/index/*" + ] }, ], }),