$ whoami --verbose role: AI Platform Manager Available

Juan Sanguineti

AI Platform Manager | Enterprise AI & LLM Systems

Building AI agent platforms that take conversational AI from prototype to production at Liberty Latin America.

Explore Projects
LinkedIn Email
5+ Years Exp.
Enterprise AI Specialization
99.9% Uptime SLA
Multi-Agent Architecture

Don't scroll. Just ask.

Tap and talk to my AI about my work.

Tap to talk

Your voice is processed in real time. No audio is stored.

Python
AWS Lambda
Amazon S3
Amazon DynamoDB
PostgreSQL
Git

Autonomous agent systems, integration pipelines, and production tools

Featured Projects & Architecture

jmsanguineti/enterprise-agent-orchestrator
Private
Enterprise · Multi-Agent

Agent Builder Framework

Framework para construir agentes conversacionales con Strands SDK + AWS AgentCore. Incluye CLI para chat local y voz bidireccional, soporte para tools/skills/steering, memory hooks, y deployment automatizado. Los agentes se despliegan a múltiples canales desde un solo codebase.

Multi-Agent Orchestration · RAG · Text-to-SQL · Stateful Graph
Python Strands SDK AWS Bedrock RAG Text-to-SQL DynamoDB AWS EKS
# Agent Builder Framework
builder = AgentBuilder.load(".")
agent-builder chat # local

System Architecture

flowchart TD
CLI["agent-builder CLI"] --> Mode{Runner?}

Mode -->|"chat"| Stream["StreamingRunner (SSE)"]
Mode -->|"voice"| Voice["VoiceRunner (WebSocket)"]
Mode -->|"async"| Async["AsyncRunner (fire-and-forget)"]

subgraph Agent["Agent Components"]
    Config["config.yml"]
    Prompt["system_prompt.md"]
    Tools["tools/ (@tool)"]
    Skills["skills/"]
end

Stream --> Agent
Voice --> Agent
Async --> Agent

Agent --> Bedrock["AWS Bedrock"]
Agent --> Memory["DynamoDB (Memory)"]
Bedrock --> AgentCore["AWS AgentCore"]

classDef default fill:#0f172a,stroke:#38bdf8,stroke-width:1px,color:#e2e8f0
classDef runner fill:#881337,stroke:#f43f5e,stroke-width:1px,color:#fecdd3
classDef component fill:#1e1b4b,stroke:#818cf8,stroke-width:1px,color:#e0e7ff
classDef aws fill:#022c22,stroke:#10b981,stroke-width:1px,color:#d1fae5

class Stream,Voice,Async runner
class Config,Prompt,Tools,Skills component
class Bedrock,Memory,AgentCore aws

Agent Builder Framework Architecture

Framework para construir agentes con Strands SDK + AWS AgentCore. Una CLI unificada (agent-builder chat|voice) que detecta automáticamente el runner según el modelo configurado. Cada agente es un directorio con config.yml, system_prompt.md, tools/, y skills/.

StreamingRunner

Modo por defecto para chat de texto. HTTP streaming con SSE para respuestas en tiempo real.

VoiceRunner

Audio bidireccional vía WebSocket. Auto-detecta modelos de voz (Nova Sonic). PCM 16-bit, 16kHz.

Tools & Skills

Decorador @tool para funciones locales. Skills como módulos reutilizables con steering handlers opcionales.

Memory Hooks

DynamoDB con TTL automático para persistir sesiones, memoria long-term por actor, y trazas de ejecución.

Deployment

Los agentes se despliegan a AWS AgentCore y se conectan a múltiples canales (WhatsApp, Telegram, Teams, web) desde un solo codebase via Channel Integrator.
jmsanguineti/smart-parser-claude-plugin
Private
Production Tooling

Channel Integrator

Orquestador que conecta canales de mensajería (WhatsApp, Telegram, Teams, web) con agentes conversacionales. Incluye transcripción automática de audio via AWS Transcribe Streaming y un flujo unificado: autenticación → validación → invocación → formateo → envío.

LLM-Powered Agentic Skills · Bedrock API · Python
Bedrock API Python Prompt Engineering Parser Engine
# Channel Integrator
class MiCanal(BaseChannel):
ChannelIntegrator(agent=arn, channel=MiCanal)

System Architecture

flowchart LR
Channels["WhatsApp / Telegram / Teams / Web"] --> Lambda["AWS Lambda"]

Lambda --> CI{"ChannelIntegrator"}

CI --> Auth["authenticate()"]
Auth --> Val["validate_input()"]
Val --> Prep["prepare_agent_input()"]
Prep --> Agent["AWS Bedrock Agent"]
Agent --> Format["format_agent_output()"]
Format --> Send["send_output()"]

Prep -->|"audio"| Transcribe["AWS Transcribe"]
Transcribe --> Agent

classDef default fill:#0f172a,stroke:#38bdf8,stroke-width:1px,color:#e2e8f0
classDef core fill:#881337,stroke:#f43f5e,stroke-width:1px,color:#fecdd3
classDef aws fill:#022c22,stroke:#10b981,stroke-width:1px,color:#d1fae5

class CI core
class Agent,Transcribe,Lambda aws

Channel Integrator Architecture

Framework que conecta canales de mensajería con agentes conversacionales. Implementás 4 métodos (validate, prepare, format, send) y el framework ejecuta el flujo completo: autenticación → validación → invocación → formateo → envío.

Multi-Channel

Un solo agente se despliega a WhatsApp, Telegram, Teams, y web. Cada canal hereda de BaseChannel.

Audio Transcription

Transcripción automática via AWS Transcribe Streaming (~2s latencia). Soporta OGG, WAV, FLAC, PCM.

Serverless

Lambda + SNS/SQS para orquestación event-driven resiliente con reintentos automáticos.

Implementation

El canal descarga audio, lo pasa como bytes, y el framework auto-detecta formato y transcribe. Ideal para bots de WhatsApp y Telegram que reciben notas de voz.
jmsanguineti/distributed-event-pipeline
Private
Cloud Data Architecture

Analytics Framework (ANA)

Framework completo para procesamiento y análisis de datos con componentes especializados: procesador multi-formato (CSV, Excel, Parquet), sincronizador de rutas S3, orquestador de flujos SQL, y procesador de consultas con Athena. Desplegado con AWS Lambda y SageMaker.

Millions of daily events · Serverless REST API Integration
Python Pandas RESTful APIs AWS S3 AWS Lambda Athena PostgreSQL
# Analytics Framework (ANA)
s3://bucket/files/mi_tabla/datos.csv
# → tabla Athena automática

System Architecture

flowchart TD
    subgraph S3["S3 Bucket (app)"]
        Files["/files/"]
        Paths["/paths/"]
        Pipelines["/pipelines/"]
        Tables["/tables/"]
    end

    Files --> LambdaFiles["Lambda Files"]
    Paths --> LambdaPaths["Lambda Paths"]
    Pipelines --> LambdaPipelines["Lambda Pipelines"]

    LambdaFiles -->|"archivos grandes"| SageMakerFiles["SageMaker Files"]
    LambdaPaths --> Files
    LambdaPipelines --> StepFn["Step Functions"]

    SageMakerFiles --> SageMakerQuery["SageMaker Query"]
    StepFn --> SageMakerQuery

    SageMakerQuery --> Glue["Glue + Lake Formation"]
    Glue --> Athena[("Athena")]

    classDef default fill:#0f172a,stroke:#38bdf8,stroke-width:1px,color:#e2e8f0
    classDef s3 fill:#1e1b4b,stroke:#818cf8,stroke-width:1px,color:#e0e7ff
    classDef lambda fill:#881337,stroke:#f43f5e,stroke-width:1px,color:#fecdd3
    classDef compute fill:#022c22,stroke:#10b981,stroke-width:1px,color:#d1fae5

    class Files,Paths,Pipelines,Tables s3
    class LambdaFiles,LambdaPaths,LambdaPipelines lambda
    class SageMakerFiles,SageMakerQuery,StepFn,Glue,Athena compute

Analytics Framework (ANA) Architecture

Framework para procesar, transformar y analizar datos en AWS. Automatiza el ciclo completo desde la ingesta hasta tablas consultables en Athena. Subís un archivo a S3 y queda como tabla automáticamente.

Files Processor

Procesa CSV, Excel, Parquet y los convierte a tablas Glue. Archivos chicos en Lambda, grandes en SageMaker.

Paths Sync

Sincroniza archivos desde otros buckets hacia /files/ basado en archivos .txt de configuración.

Pipelines Orchestrator

Ejecuta queries SQL programadas (diario, semanal) con Step Functions para garantizar orden secuencial.

Query Processor

Contenedor SageMaker que ejecuta queries en Athena y guarda resultados como Parquet particionado. Integrado con Glue Catalog y Lake Formation para governance.
jmsanguineti/portfolio
main
Live demo

Voice AI Agent for Portfolio

Real-time conversational voice agent embedded in this site. Visitors can talk to an AI that answers questions about my projects, stack and experience, with a state-driven interface (idle, listening, thinking, speaking) and audio processed by Google AgentCore.

AgentCore Live API · State Machine · Audio Context
AgentCore Web Audio API WebSockets CDK
# Real-time Web Audio API & AgentCore WS
this.ws = new WebSocket(geminiWsUrl)
this.mic = new MicCapture(ctx, sendAudioChunk)
Architecture Try it live

Technologies I work with

Tech Stack

01.AI & GenAI Platform Management

10 tools
AWS Bedrock Strands SDK LangChain RAG Text-to-SQL Multi-Agent Orchestration AgentCore API OpenAI API Anthropic Bedrock API Vector Databases

02.Cloud & Infrastructure

8 tools
AWS Lambda AWS S3 API Gateway DynamoDB Amazon SQS SageMaker Serverless Event-Driven

03.Data & Languages

7 tools
Python SQL Athena PostgreSQL ETL/ELT Pipelines RESTful APIs TypeScript

04.Tools & DevOps

7 tools
Git Docker CI/CD AWS Glue SFTP CDK Docker

My professional journey

Experience

AI Platform Manager

Liberty Latin America

commit 2025..Present
  • Architected a stateful multi-agent system coordinating specialized agents (ETL Assistant, Support, Troubleshooting) via AWS Bedrock and Strands Agents SDK, enabling autonomous resolution of enterprise-level support queries.
  • Built a Document Intelligence and RAG pipeline for semantic retrieval from technical documentation, enabling non-technical teams to access specialized knowledge through natural language queries.
  • Designed and implemented a Text-to-SQL agent with schema validation and strict read-only security controls, providing safe, contextual access to relational databases.
  • Developed a LLM-powered Channel Integrator with agentic skills to generate resilient parsers and validation scripts for complex structured and semi-structured formats.
  • Designed and built scalable ML pipelines integrating data from heterogeneous sources (AWS S3, Glue, Athena, Redshift) into SageMaker, processing millions of records with automated orchestration and 99.9% uptime.
  • Implemented multi-channel integration layer deploying single agent codebases to WhatsApp, Telegram, Teams, and web with event-driven architectures (AWS Lambda, SNS, SQS) for resilient message orchestration.
  • Built serverless architectures on AWS (Lambda, S3, API Gateway, DynamoDB, SQS, Bedrock, AgentCore) for scalable, event-driven AI agent pipelines with automated evaluation.
  • Collaborated directly with enterprise clients to design integration solutions aligned with business objectives, ensuring delivery reliability and system observability.
Python AWS Bedrock Strands SDK RAG AWS SageMaker Athena PostgreSQL Glue

Data Engineering Manager

Liberty Latin America

Jan 2025 - Mar 2026
  • Led Data Engineering team responsible for ML pipelines, data architecture, and analytics infrastructure across the organization.
  • Design and optimization of automated ML pipelines using AWS SageMaker and modern data stack (Glue, Spark, Athena).
AWS SageMaker Glue Athena Spark

Senior Data Engineer

Liberty Latin America

Oct 2021 - Jan 2025
  • Designed and optimized automated ML pipelines integrated within a modern AWS data stack.
  • Data processing with AWS Glue, Spark SQL, Athena, S3, and SageMaker pipelines with containerized components (ECR).
Python AWS Glue Spark CDK

Data Engineer

Accenture

Mar 2021 - Oct 2021
  • Led migration of on-premise data pipelines to AWS for a USA-based enterprise client.
AWS EMR Hive Redshift

Data Engineer

ITE Consult

Sep 2020 - Mar 2021
  • Designed and implemented ELT program comparing internal demand planning simulations against client ERP data with SAP HANA Cloud.
Python SAP HANA REST APIs
✦ About

I engineer intelligent systems.

Senior data engineer with a strong foundation in ML pipelines, data architecture, and cloud infrastructure — now leading the AI Platform team building production-grade conversational agents with AWS Bedrock, AgentCore, and multi-channel orchestration.

I design and implement multi-agent systems using AWS Bedrock and Strands Agents SDK, RAG pipelines for semantic document retrieval, and automated evaluation pipelines. My unconventional background — philosophy (logic, epistemology) + business strategy (Strathclyde MBA) + self-taught Python and cloud architecture — shapes how I approach AI systems: I care about what the agent actually accomplishes for the user, not just whether it technically works.

My experience in enterprise integrations is not a separate chapter — it is the foundation that makes AI systems production-ready: reliable, scalable, and connected to the data sources that matter.

View Resume

// engineering evolution

Enterprise Systems
Integrations & APIs
Data Engineering Manager
Distributed Systems
AI / GenAI Systems

Previously at

Accenture & ITE Consult

Location

Argentina

Education

Philosophy & MBA — Universidad Nacional del Litoral & Strathclyde

commit 2021..2025
"The most reliable AI systems are built on solid engineering foundations — connected to real data, designed for production, and grounded in the systems they automate."
Juan Sanguineti · AI Platform Manager | Enterprise AI & LLM Systems

Let's connect

Get in Touch

$ init --collaborate

Let's build something intelligent

Open to opportunities in AI Platform Management, GenAI, and LLM Systems.

Get in Touch
or find me at

jmsanguineti@hotmail.com

>