ExploringEvidence: Medium50/100

Boti AI assistant for Buenos Aires government procedures with Amazon Bedrock and LangGraph

Use case typeConversational assistantsUpdated Jun 13, 2026

The Government of the City of Buenos Aires introduced Boti, a WhatsApp-based AI assistant, to help citizens access city information and government procedures. To better answer questions about more than 1,300 procedures, the city and AWS built an agentic AI system with Amazon Bedrock, Bedrock Knowledge Bases, and LangGraph.

Organization
Government of the City of Buenos Aires
Industry
Public Sector
Location
Argentina
Published
August 2025

Reported outcomes

12.5-17.5%

retrieval improvement over standard RAGOther quantified impact

3,000,000 conversations/monthmonthly conversations100%harmful queries blocked98.9%top-1 retrieval accuracy98%voseo usage accuracy92%periphrastic future usage accuracy

Strategic outcomes

Customer experience & trustFaster conversational access to government proceduresRisk & complianceBlocked harmful and prompt-injection contentNew product / capabilityBuilt an AI government procedures assistantBetter decisions & insightImproved procedure retrieval accuracy
Why do we believe this?Outcome claims, sources, and evidence checks

Normalized claim

Monthly conversations: 3,000,000 conversations/month increase

AWS Machine Learning BlogAug 28, 2025Blog postExplicit claimMedium evidence strength

facilitates more than 3 million conversations each month

Normalized claim

Harmful queries blocked: 100%

AWS Machine Learning BlogAug 28, 2025Blog postExplicit claimMedium evidence strength

successfully blocked 100% of harmful queries

Normalized claim

Top-1 retrieval accuracy: 98.9%

AWS Machine Learning BlogAug 28, 2025Blog postExplicit claimMedium evidence strength

achieving up to 98.9% top-1 retrieval accuracy

Normalized claim

Retrieval improvement over standard RAG: 12.5-17.5% increase

AWS Machine Learning BlogAug 28, 2025Blog postExplicit claimMedium evidence strength

marks a 12.5–17.5% improvement over standard retrieval-augmented generation (RAG) methods

Normalized claim

Voseo usage accuracy: 98%

AWS Machine Learning BlogAug 28, 2025Blog postExplicit claimMedium evidence strength

98% accurate in voseo usage

Normalized claim

Periphrastic future usage accuracy: 92%

AWS Machine Learning BlogAug 28, 2025Blog postExplicit claimMedium evidence strength

92% accurate in periphrastic future usage

Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks
Customer
Government of the City of Buenos Aires
Provider
AWS
Maturity
Exploring

The guardrail system blocked 100% of harmful queries in evaluation

Customer identity supportedSource describes one deploymentMaturity supported

Primary read

Use case focus

Showing 3 of 4

  • 1Citizen Services
  • 2Government Assistant
  • 3Conversational AI
  • Citizens needed faster, more conversational access to complex government procedures across more than 1,300 rules, exceptions, and variants.
  • The city also needed to block harmful content and prompt-injection attempts while keeping responses useful and on-brand in Rioplatense Spanish.
  • The solution uses an input guardrail system with a custom LLM classifier that approves or blocks requests.
  • Approved requests are handled by a government procedures agent that retrieves relevant information from Amazon Bedrock Knowledge Bases and generates answers with the Amazon Bedrock Converse API.
  • The team built a reasoning retriever that uses comparative summaries and optional LLM reasoning to disambiguate similar procedures.
  • Responses are prompted to match Boti's characteristic style, including voseo and local phrasing.
Technologies
Amazon BedrockAmazon Bedrock Knowledge BasesAmazon Bedrock Converse APIAmazon Titan Text Embeddings V2LangGraph
  • Boti handles more than 3 million conversations per month.
  • The guardrail system blocked 100% of harmful queries in evaluation.
  • The reasoning retriever achieved up to 98.9% top-1 retrieval accuracy, a 12.5-17.5% improvement over standard RAG methods.
  • Subject matter experts rated the responses 98% accurate for voseo usage and 92% accurate for periphrastic future usage.
Architecture

The architecture combines a parallel input-guardrail and agent workflow orchestrated in LangGraph. The guardrail uses a custom LLM classifier to approve or block requests. The agent retrieves documents and metadata from Amazon Bedrock Knowledge Bases, uses the Amazon Bedrock Converse API for inference, and applies a selective reasoning-retrieval step based on comparative summaries to disambiguate similar government procedures before generating Rioplatense Spanish responses.

Sources & evidence1
Evidence: Medium50/100Evidence strength
  • Customer explicitly identified
  • Deployment status explicitly supported
  • Quantified outcome available
  • Technical implementation details available
Type: Blog PostPublished: Aug 28, 2025Publisher: AWSEvidence: VendorConfidence: Medium
Primary source

AI-generated summary. Verify important details with the linked sources before relying on this case.

Explore related AI use cases
Customer evidence
More from Government of the City of Buenos Aires
Provider evidence
AWS provider pageAWS ranked cases
Market context
Public Sector AI use casesPublic Sector industry insightsArgentina country insights
AI patterns
AgentRAG
Browse the catalog
All AI use casesAI use-case typesConversational assistants use cases

Was this useful?

Community

Comments

No published comments yet.