Innovation and technology nurture each other.

We harness the power of data for better decision making within an innovative organizational culture, through our team of Data Architects, Data Scientists and Data Engineers.

This evolves into gigantic volumes of data, which using the tools provided by the cloud allows us not only to improve its analysis but also to implement Machine Learning and AI to increase business intelligence.

Intelligent Forecasting

AI solution that allows the forecasting of business events in advance in order to plan improvements and actions that optimize results.

Data Lakes

Business analysis, manage diverse data sources, and achieve a better understanding of the world through these centralized repositories.

Master Data Optimization

The Master Data Optimization solution uses Machine Learning to automate the creation, remediation and continuous maintenance of master databases.

HealthBot

Virtual health channel that uses cognitive technologies such as artificial intelligence, in order to optimize and speed up the times of health institutions.

Generative AI

Discover the technological disruption that allows you to take your business to a new level and achieve maximum productivity.

Intelligent Automation

Robotic Process Automation (RPA) and Artificial Intelligence that empower a rapid automation of end-to-end business processes and accelerate digital transformation.

Nubiral Cognitive AI Bot

Virtual conversational assistant based on Artificial Intelligence (AI) that enables real-time file processing.

Intelligent Document Processing

An AI solution that allows extracting information from documents and incorporating it into an automatic process, using OCR technology.

Expert help to manage infrastructure and data

This important Argentine energy company chose Nubiral to obtain high-level support for its Oracle solutions and an advanced monitoring system for its critical IT assets.

Read more

Optimal infrastructure monitoring with Zabbix

A single platform and a single visualization to gain efficiency when managing the more than 1,600 devices of this energy company with a solid presence in Latin America and more than 100 years in the industry.

Read more

Modernizing DevOps to take agility to the next level

This leading Colombian company in the hydrocarbon transportation and logistics industry is now able to respond quickly to the demands of a fast-moving market.

Read more

New data platform with analytics and machine learning

The implementation of Microsoft Fabric enables this company in charge of coordinating the generation, transmission and distribution of power in Argentina to capitalize on the value of its data and even generate new monetization alternatives.

Read more
Blog

Microsoft Fabric on Azure: Best practices for migration

The successful and careful migration of data to Microsoft Fabric on the Azure platform allows maximizing the use of data for business benefit.

Read more
eBooks

Banking & Fintech: How to integrate GenAI in fraud detection

A guide for financial firms to discover the power of this new technology to optimize their fraud detection strategies.

Read more
Papers

Banking and Fintech: How to get value from emerging technologies?

A guide for companies to start capitalizing on their investments in new technologies now.

Read more
Whitepapers

Machine learning recommender systems in digital media companies

Advances in machine learning enable digital media companies to improve their recommender systems and optimize user experience.

Read more

Learn about our architecture that combines at least one public and one private cloud to deliver the highest levels of scalability, flexibility, and performance.

Deployment of AWS Control Tower and migration of services to Openshift

A major bank needed to migrate its workloads to the cloud and decided to rely on Nubiral for the initial configuration in AWS.

Read more

Application modernization by migrating to the AWS cloud

Migration to the AWS Cloud at Telecom Argentina, modernizing obsolete applications with a focus on operational excellence.

Read more

Modernization of multimedia content with AWS Migration

Successful migration to AWS cloud to modernize Claro Video’s multimedia content infrastructure.

Read more

Migration to AWS by a major Japanese automobile company

The smooth migration of Toyota to AWS unleashes performance, cost efficiency, and user satisfaction.

Read more
Blog

EC2 & AWS: Powering business in the cloud

A tool that offers flexibility, scalability and efficiency, and could become the keystone to take your business to a new level.

Read more
eBooks

Cloud 4.0: A phenomenon in exponential growth

A tour of the main opportunities that arise from a correct and timely migration of workloads to the cloud, and the trends that are being visualized in the cloud universe.

Read more
Papers
04 October , 2022

Cloud 4.0: A phenomenon in exponential growth

A tour of the main opportunities that arise from a correct and timely migration of workloads to the cloud, and the trends that are being visualized in the cloud universe.

Read more

Application migration, optimization, efficiency, security, analytics and implementation services, with the aim of simplifying and accelerating the adoption of the latest IT trends in the market.

Planning and Consulting

  • Evaluation and Planning
  • Adoption Strategy Consultancy

Test

  • Test Automation

Execution

  • Design and implementation of CI/CD Pipeline
  • Automation and implementation of processes

App Dev

  • Code (JavaScript, Go, Python)

Development to integrate Gala chatbot into the CloudGuru educational platform

The client needed to evolve their Gala chatbot so that end users could consume information from their centralized documentation platform.

Read more

Migration of CI/CD to Github

Important bank in Colombia migrates from GitLab to GitHub Enterprise and trains its staff for efficient adoption.

Read more

Migration of CI/CD to Github

Migration and Configuration of GitHub Enterprise Server for a major financial group, focused on modernizing its on-premise CI/CD structure.

Read more

Telecommunications modernization with AWS technologies

A leading telecommunications company modernizes its applications to respond more quickly and agilely to market changes.

Read more
Blog

GitHub: Features for fintech web application development

This powerful tool constitutes a complete ecosystem that boosts efficiency and collaboration in all aspects of the software development cycle.

Read more
eBooks

Agile & DevOps

A review of the meanings of each of these concepts, how they integrate with each other and what benefits they bring.

Read more

We scale to environments with thousands of items monitored simultaneously.
We also capture data of systems and applications over time to make proactive decisions and to anticipate disruptions in business services.

  • Zabbix Architecture and Implementation
  • Data & Analytics Monitoring

Implementation of monitoring solution with Zabbix

Private bank implements comprehensive Observability solution with Nubiral to optimize and gain greater visibility of its Infrastructure health.

Read more

Implementation of OpenSearch

Improving data observability and security at a major bank with AWS OpenSearch.

Read more

Implementation of OpenSearch

One of the leading banks in Chile utilizes the most advanced AWS services to work on the ingestion, storage, detection, and predictive models of data from cybersecurity intelligence sources.

Read more

Monitoring solution upgrade using Zabbix

Migration of monitoring tool to an automated system for host discovery, dashboards, and scalability over time.

Read more
Blog

Why should companies implement observability solutions?

In addition to anticipating and preventing IT infrastructure problems that impact the business, this strategy is key to boosting the user experience.

Read more
eBooks

Compliance: the evolution of monitoring

A key paradigm for anticipating and solving problems in increasingly complex IT infrastructures.

Read more
Whitepapers

OpenSearch and its log agents

OpenSearch is a comprehensive solution for centralizing and analyzing logs from various sources, ideal for managing complex IT scenarios.

Read more

We help innovate by preparing our clients against current cyber threats.

We fulfill the responsibility of protecting data to maintain trust and comply with regulations.

  • Cloud Assessment & Consulting
  • Security Frameworks & Best Practices
  • Penetration Testing
  • Cloud Security
  • Governance, Risk & Compliance
  • DevSecOps
Blog

Cybersecurity: A key pillar for a 360° digital experience

Mitigating risks linked to cyberattacks and protecting data is essential to survive and lead in the current era of digital transformation.

Read more
Whitepapers

Cybersecurity in your company: The 360º digital solution from Nubiral

How to develop a cybersecurity plan? Which are the main threats? Which are the best and most modern technologies to face these threats?

Read more

Microsoft Fabric on Azure: Best practices for migration

The successful and careful migration of data to Microsoft Fabric on the Azure platform allows maximizing the use of data for business benefit.

Read more

Why should companies implement observability solutions?

In addition to anticipating and preventing IT infrastructure problems that impact the business, this strategy is key to boosting the user experience.

Read more

Generative AI in Oil & Gas: 5 highly complex use cases

Key applications of this new technology that contribute sustainably to the progress of the sector.

Read more

Data augmentation in images: Uses and benefits

The strategy of creating synthetic data, known as data augmentation in images, is key to obtaining the maximum added value from computer vision.

Read more

Banking & Fintech: How to integrate GenAI in fraud detection

A guide for financial firms to discover the power of this new technology to optimize their fraud detection strategies.

Read more

Transform your business with the power of Azure OpenAI Service

How to quickly and easily create your own Copilot and Generative AI applications.

Read more

GenAI use cases with Amazon Bedrock

Discover the potential of digital transformation with Generative AI.

Read more

MLOps: powering the value of machine learning

A comprehensive guide to MLOps, a key discipline that guarantees the success of Machine Learning (ML) projects in organizations.

Read more

Connect+ is a great tool to incorporate knowledge and stay up to date with the latest technological developments.

Access new innovative audiovisual content, quickly and easily. Explore and get to know the technological universe in a different and agile way!

Machine learning recommender systems in digital media companies

Advances in machine learning enable digital media companies to improve their recommender systems and optimize user experience.

Read more

Cybersecurity in your company: The 360º digital solution from Nubiral

How to develop a cybersecurity plan? Which are the main threats? Which are the best and most modern technologies to face these threats?

Read more

Microsoft Fabric Guide: Use case end-to-end Deployment

Banks and financial services companies can benefit in numerous ways by deploying Microsoft Fabric.

Read more

How to Deploy Microsoft Fabric in Multicloud Infrastructures

Microsoft Fabric’s data analytics combined with the power of the multi-cloud architecture, drives decision making and empowers users.

Read more

AWS + Nubiral

As an Advanced Consulting Partner of the AWS Partner Network, we think outside the box, daring to go where no one has gone before.
We constantly challenge ourselves to be better, providing your company with AWS solutions in a holistic and tailored way.

Microsoft + Nubiral

As Cloud Gold Partner, we work together with Microsoft every day to offer our clients the most innovative solutions based on the different microservices and capabilities that the Azure cloud offers.
Our team is constantly training and certifying on Azure’s services.

Zabbix + Nubiral

We scale to environments with thousands of items monitored simultaneously.
We also capture data of systems and applications over time to make proactive decisions and to anticipate disruptions in business services.

Diligent + Nubiral

Nubiral and Diligent join to present a revolutionary solution that will transform the way organizations manage their GRC (Governance, Risk, and Compliance) processes.

Data
& Innovation

General Info

Innovation and technology nurture each other.

We harness the power of data for better decision making within an innovative organizational culture, through our team of Data Architects, Data Scientists and Data Engineers.

This evolves into gigantic volumes of data, which using the tools provided by the cloud allows us not only to improve its analysis but also to implement Machine Learning and AI to increase business intelligence.

Solutions

Intelligent Forecasting

AI solution that allows the forecasting of business events in advance in order to plan improvements and actions that optimize results.

Data Lakes

Business analysis, manage diverse data sources, and achieve a better understanding of the world through these centralized repositories.

Master Data Optimization

The Master Data Optimization solution uses Machine Learning to automate the creation, remediation and continuous maintenance of master databases.

HealthBot

Virtual health channel that uses cognitive technologies such as artificial intelligence, in order to optimize and speed up the times of health institutions.

Generative AI

Discover the technological disruption that allows you to take your business to a new level and achieve maximum productivity.

Intelligent Automation

Robotic Process Automation (RPA) and Artificial Intelligence that empower a rapid automation of end-to-end business processes and accelerate digital transformation.

Nubiral Cognitive AI Bot

Virtual conversational assistant based on Artificial Intelligence (AI) that enables real-time file processing.

Intelligent Document Processing

An AI solution that allows extracting information from documents and incorporating it into an automatic process, using OCR technology.

Success Stories

Expert help to manage infrastructure and data

This important Argentine energy company chose Nubiral to obtain high-level support for its Oracle solutions and an advanced monitoring system for its critical IT assets.

Read more

Optimal infrastructure monitoring with Zabbix

A single platform and a single visualization to gain efficiency when managing the more than 1,600 devices of this energy company with a solid presence in Latin America and more than 100 years in the industry.

Read more

Modernizing DevOps to take agility to the next level

This leading Colombian company in the hydrocarbon transportation and logistics industry is now able to respond quickly to the demands of a fast-moving market.

Read more

New data platform with analytics and machine learning

The implementation of Microsoft Fabric enables this company in charge of coordinating the generation, transmission and distribution of power in Argentina to capitalize on the value of its data and even generate new monetization alternatives.

Read more

Connect

Blog

Microsoft Fabric on Azure: Best practices for migration

The successful and careful migration of data to Microsoft Fabric on the Azure platform allows maximizing the use of data for business benefit.

Read more
eBooks

Banking & Fintech: How to integrate GenAI in fraud detection

A guide for financial firms to discover the power of this new technology to optimize their fraud detection strategies.

Read more
Papers

Banking and Fintech: How to get value from emerging technologies?

A guide for companies to start capitalizing on their investments in new technologies now.

Read more
Whitepapers

Machine learning recommender systems in digital media companies

Advances in machine learning enable digital media companies to improve their recommender systems and optimize user experience.

Read more

Hybrid
Multi Cloud

General Info

Learn about our architecture that combines at least one public and one private cloud to deliver the highest levels of scalability, flexibility, and performance.

Success Stories

Deployment of AWS Control Tower and migration of services to Openshift

A major bank needed to migrate its workloads to the cloud and decided to rely on Nubiral for the initial configuration in AWS.

Read more

Application modernization by migrating to the AWS cloud

Migration to the AWS Cloud at Telecom Argentina, modernizing obsolete applications with a focus on operational excellence.

Read more

Modernization of multimedia content with AWS Migration

Successful migration to AWS cloud to modernize Claro Video’s multimedia content infrastructure.

Read more

Migration to AWS by a major Japanese automobile company

The smooth migration of Toyota to AWS unleashes performance, cost efficiency, and user satisfaction.

Read more

Connect

Blog

EC2 & AWS: Powering business in the cloud

A tool that offers flexibility, scalability and efficiency, and could become the keystone to take your business to a new level.

Read more
eBooks

Cloud 4.0: A phenomenon in exponential growth

A tour of the main opportunities that arise from a correct and timely migration of workloads to the cloud, and the trends that are being visualized in the cloud universe.

Read more
Papers
04 October , 2022

Cloud 4.0: A phenomenon in exponential growth

A tour of the main opportunities that arise from a correct and timely migration of workloads to the cloud, and the trends that are being visualized in the cloud universe.

Read more

DevOps
& App Evolution

General Info

Application migration, optimization, efficiency, security, analytics and implementation services, with the aim of simplifying and accelerating the adoption of the latest IT trends in the market.

Solutions

Planning and Consulting

  • Evaluation and Planning
  • Adoption Strategy Consultancy

Test

  • Test Automation

Execution

  • Design and implementation of CI/CD Pipeline
  • Automation and implementation of processes

App Dev

  • Code (JavaScript, Go, Python)

Success Stories

Development to integrate Gala chatbot into the CloudGuru educational platform

The client needed to evolve their Gala chatbot so that end users could consume information from their centralized documentation platform.

Read more

Migration of CI/CD to Github

Important bank in Colombia migrates from GitLab to GitHub Enterprise and trains its staff for efficient adoption.

Read more

Migration of CI/CD to Github

Migration and Configuration of GitHub Enterprise Server for a major financial group, focused on modernizing its on-premise CI/CD structure.

Read more

Telecommunications modernization with AWS technologies

A leading telecommunications company modernizes its applications to respond more quickly and agilely to market changes.

Read more

Connect

Blog

GitHub: Features for fintech web application development

This powerful tool constitutes a complete ecosystem that boosts efficiency and collaboration in all aspects of the software development cycle.

Read more
eBooks

Agile & DevOps

A review of the meanings of each of these concepts, how they integrate with each other and what benefits they bring.

Read more

Monitoring
& Intelligence

General Info

We scale to environments with thousands of items monitored simultaneously.
We also capture data of systems and applications over time to make proactive decisions and to anticipate disruptions in business services.

Solutions

  • Zabbix Architecture and Implementation
  • Data & Analytics Monitoring

Success Stories

Implementation of monitoring solution with Zabbix

Private bank implements comprehensive Observability solution with Nubiral to optimize and gain greater visibility of its Infrastructure health.

Read more

Implementation of OpenSearch

Improving data observability and security at a major bank with AWS OpenSearch.

Read more

Implementation of OpenSearch

One of the leading banks in Chile utilizes the most advanced AWS services to work on the ingestion, storage, detection, and predictive models of data from cybersecurity intelligence sources.

Read more

Monitoring solution upgrade using Zabbix

Migration of monitoring tool to an automated system for host discovery, dashboards, and scalability over time.

Read more

Connect

Blog

Why should companies implement observability solutions?

In addition to anticipating and preventing IT infrastructure problems that impact the business, this strategy is key to boosting the user experience.

Read more
eBooks

Compliance: the evolution of monitoring

A key paradigm for anticipating and solving problems in increasingly complex IT infrastructures.

Read more
Whitepapers

OpenSearch and its log agents

OpenSearch is a comprehensive solution for centralizing and analyzing logs from various sources, ideal for managing complex IT scenarios.

Read more

Cybersecurity

General Info

We help innovate by preparing our clients against current cyber threats.

We fulfill the responsibility of protecting data to maintain trust and comply with regulations.

Solutions

  • Cloud Assessment & Consulting
  • Security Frameworks & Best Practices
  • Penetration Testing
  • Cloud Security
  • Governance, Risk & Compliance
  • DevSecOps

Connect

Blog

Cybersecurity: A key pillar for a 360° digital experience

Mitigating risks linked to cyberattacks and protecting data is essential to survive and lead in the current era of digital transformation.

Read more
Whitepapers

Cybersecurity in your company: The 360º digital solution from Nubiral

How to develop a cybersecurity plan? Which are the main threats? Which are the best and most modern technologies to face these threats?

Read more

Partners

Solutions

AWS + Nubiral

As an Advanced Consulting Partner of the AWS Partner Network, we think outside the box, daring to go where no one has gone before.
We constantly challenge ourselves to be better, providing your company with AWS solutions in a holistic and tailored way.

Microsoft + Nubiral

As Cloud Gold Partner, we work together with Microsoft every day to offer our clients the most innovative solutions based on the different microservices and capabilities that the Azure cloud offers.
Our team is constantly training and certifying on Azure’s services.

Zabbix + Nubiral

We scale to environments with thousands of items monitored simultaneously.
We also capture data of systems and applications over time to make proactive decisions and to anticipate disruptions in business services.

Diligent + Nubiral

Nubiral and Diligent join to present a revolutionary solution that will transform the way organizations manage their GRC (Governance, Risk, and Compliance) processes.

Success Stories

Innovative Blockchain solution with Smart Contracts on AWS

The project focuses on implementing a private blockchain platform to improve transparency, security, and efficiency in deployment and development processes.

Read more

A medical center implements a chatbot and cognitive services

Improvement in patient care times and reduction in administrative staff dedication costs for routine tasks.

Read more

Implementation of monitoring solution with Zabbix

Private bank implements comprehensive Observability solution with Nubiral to optimize and gain greater visibility of its Infrastructure health.

Read more

Connect

Blog

Microsoft Fabric on Azure: Best practices for migration

The successful and careful migration of data to Microsoft Fabric on the Azure platform allows maximizing the use of data for business benefit.

Read more

Why should companies implement observability solutions?

In addition to anticipating and preventing IT infrastructure problems that impact the business, this strategy is key to boosting the user experience.

Read more

Generative AI in Oil & Gas: 5 highly complex use cases

Key applications of this new technology that contribute sustainably to the progress of the sector.

Read more

Data augmentation in images: Uses and benefits

The strategy of creating synthetic data, known as data augmentation in images, is key to obtaining the maximum added value from computer vision.

Read more

eBooks & Papers

Banking & Fintech: How to integrate GenAI in fraud detection

A guide for financial firms to discover the power of this new technology to optimize their fraud detection strategies.

Read more

Transform your business with the power of Azure OpenAI Service

How to quickly and easily create your own Copilot and Generative AI applications.

Read more

GenAI use cases with Amazon Bedrock

Discover the potential of digital transformation with Generative AI.

Read more

MLOps: powering the value of machine learning

A comprehensive guide to MLOps, a key discipline that guarantees the success of Machine Learning (ML) projects in organizations.

Read more

Connect+

Connect+ is a great tool to incorporate knowledge and stay up to date with the latest technological developments.

Access new innovative audiovisual content, quickly and easily. Explore and get to know the technological universe in a different and agile way!

Whitepapers

Machine learning recommender systems in digital media companies

Advances in machine learning enable digital media companies to improve their recommender systems and optimize user experience.

Read more

Cybersecurity in your company: The 360º digital solution from Nubiral

How to develop a cybersecurity plan? Which are the main threats? Which are the best and most modern technologies to face these threats?

Read more

Microsoft Fabric Guide: Use case end-to-end Deployment

Banks and financial services companies can benefit in numerous ways by deploying Microsoft Fabric.

Read more

How to Deploy Microsoft Fabric in Multicloud Infrastructures

Microsoft Fabric’s data analytics combined with the power of the multi-cloud architecture, drives decision making and empowers users.

Read more
Guías Técnicas

Asistente conversacional basado en IA

Paso a paso, desde la toma de requerimientos hasta la mejora continua, cómo desarrollar un asistente conversacional basado en inteligencia artificial.

Home / asistente conversacional virtual

1. Por qué son tan importantes los asistentes conversacionales basados en IA

En un mundo donde las experiencias automatizadas, coherentes y de alto nivel son esenciales para el éxito empresarial, la importancia de los asistentes conversacionales basados en IA es innegable.

Desde la identificación de requerimientos y objetivos hasta la iteración para lograr una mejora continua, aquí presentamos un detallado paso a paso para construir estas herramientas empresariales clave.

2. Guía de desarrollo

1. Identificación de requerimientos y objetivos

Antes de sumergirse en la técnica, es esencial definir el propósito, el tipo de preguntas que debe responder y el ámbito de aplicación del asistente virtual. Lo mismo con las fuentes de información que le vamos a brindar para que cuente con el conocimiento necesario y dé respuestas adecuadas. Y, por supuesto, cuál será el canal en que el asistente esté disponible para los usuarios. Puede ser un frontend o estar integrado al chat (Slack, Whatsapp, Microsoft Teams, etc).

 

2. Recolección y preprocesamiento de datos

Pueden ser bases de datos estructuradas, no estructuradas o archivos planos como PDFs. Lo importante es detectar dónde están las fuentes de datos que permitirán dar contexto al asistente. A partir de ahí, definimos el procesamiento que debemos hacerle.

 

3. Construcción del embedding

Los embeddings son una colección de vectores que capturan la esencia del contenido. Esta transformación permite que el conocimiento que está en los archivos sea accesible y utilizable en tiempo real por los usuarios. Existen distintos modelos de embeddings, como Microsoft Ada o Amazon Titan. En este paso seleccionamos también el que se adecue más a nuestros requerimientos.

 

4. Creación de una base de datos vectorial

Una base de datos vectorial especializada como Azure Cognitive Search o Amazon OpenSearch para almacenar y recuperar embeddings permite buscar rápidamente respuestas o sugerencias basadas en similitud semántica.

 

5. Configuración y ajuste del LLM

Uno de los pasos más importantes. Diseñar prompts que guíen al modelo a generar respuestas alineadas con el propósito del asistente. Para eso, se debe experimentar con diferentes modelos de LLM, analizar cuál se ajusta mejor a las necesidades y configurar el prompt para que el asistente se comporte de forma asertiva. En algunos casos se requiere un fine-tunning para ajustar el modelo LLM para la tarea definida.

 

6. Integración del modelo con una API

Para que el asistente sea accesible y se integre en diferentes plataformas, hay que envolver el modelo LLM en una API, con herramientas como FastAPI, Flask o Django. Esta debería poder recibir un input del usuario, procesarlo a través del modelo y devolver una respuesta. Luego, puede desplegarse en un entorno cloud para lograr una aplicación segura y escalable.

 

7. Desarrollo del interfaz e integración de la aplicación

Si el caso de uso lo requiere, desarrollar una interfaz de usuario (UI) amigable y funcional para que los usuarios interactúen con el asistente. Puede ser una aplicación web, móvil o integrada en otro software. Luego debemos realizar la integración necesaria con las APIs de backend para que cobre vida.

 

8. Monitoreo y alertas

Las herramientas de monitoreo y logging rastrean las interacciones del usuario y las respuestas del asistente.
Se definen métricas clave de rendimiento (
KPIs) y establecen umbrales para alertas. Por ejemplo, si el asistente no puede responder a un cierto porcentaje de preguntas.

 

9. Iteración y mejora continua

Los datos recolectados a través del monitoreo permiten identificar áreas de mejora. También se podría solicitar retroalimentación a los usuarios y a partir de este feedback realizar ajustes.
Esto puede incluir ajustar el prompt, afinar el embedding o expandir la base de datos con nuevas fuentes de información.
Estos pasos dan una estructura sólida para crear, implementar y mantener un asistente conversacional impulsado por IA.

3. Conclusiones

– El primer paso antes de construir un asistente conversacional basado en IA es entender qué valor aportará al negocio.

– Con eso resuelto, se avanzará sobre los requerimientos y los objetivos puntuales del proyecto y con la identificación de las fuentes de datos que le darán contexto al asistente.

– Luego se construyen embeddings para interpretar el contenido de los documentos y bases de datos vectoriales para buscar respuestas o sugerencias basadas en similitud semántica.

– El siguiente paso es esencial: diseñar los prompts que guíen al modelo a generar respuestas alineadas con el propósito. Hecho esto, se utilizan API para que el modelo esté disponible para los usuarios.

– El desarrollo de una interfaz amigable es la clave para una mayor adopción. No obstante, gracias al monitoreo y el logging es posible verificar si el funcionamiento del asistente se da de acuerdo a lo esperado.

– Y este es apenas el principio del recorrido: siempre existen oportunidades para mejorar el asistente.

Asistente conversacional basado en IA: guía de desarrollo

 

1. Introducción: por qué son tan importantes los asistentes conversacionales basados en IA

Desde la identificación de requerimientos y objetivos hasta la iteración para lograr una mejora continua, un paso a paso para construir este tipo de herramientas.

Los asistentes conversacionales basados en IA se vuelven cada vez más atractivos para las empresas que desean brindar experiencias automatizadas, coherentes y de alto nivel. Aquí, los pasos para desarrollar uno.

 

2. Identificación de requerimientos y objetivos

Antes de sumergirse en la técnica, es esencial definir el propósito, el tipo de preguntas que debe responder y el ámbito de aplicación del asistente virtual. Lo mismo con las fuentes de información que le vamos a brindar para que cuente con el conocimiento necesario y dé respuestas adecuadas. Y, por supuesto, cuál será el canal en que el asistente esté disponible para los usuarios. Puede ser un frontend o estar integrado al chat (Slack, Whatsapp, Microsoft Teams, etc).

 

3. Recolección y preprocesamiento de datos

Pueden ser bases de datos estructuradas, no estructuradas o archivos planos como PDFs. Lo importante es detectar dónde están las fuentes de datos que permitirán dar contexto al asistente. A partir de ahí, definimos el procesamiento que debemos hacerle.

 

4. Construcción del embedding

Los embeddings son una colección de vectores que capturan la esencia del contenido. Esta transformación permite que el conocimiento que está en los archivos sea accesible y utilizable en tiempo real por los usuarios. Existen distintos modelos de embeddings, como Microsoft Ada o Amazon Titan. En este paso seleccionamos también el que se adecue más a nuestros requerimientos.

 

5. Creación de una base de datos vectorial

Una base de datos vectorial especializada como Azure Cognitive Search o Amazon OpenSearch para almacenar y recuperar embeddings permite buscar rápidamente respuestas o sugerencias basadas en similitud semántica.

 

6. Configuración y ajuste del LLM

Uno de los pasos más importantes. Diseñar prompts que guíen al modelo a generar respuestas alineadas con el propósito del asistente. Para eso, se debe experimentar con diferentes modelos de LLM, analizar cuál se ajusta mejor a las necesidades y configurar el prompt para que el asistente se comporte de forma asertiva. En algunos casos se requiere un fine-tunning para ajustar el modelo LLM para la tarea definida.

 

7. Integración del modelo con una API

Para que el asistente sea accesible y se integre en diferentes plataformas, hay que envolver el modelo LLM en una API, con herramientas como FastAPI, Flask o Django. Esta debería poder recibir un input del usuario, procesarlo a través del modelo y devolver una respuesta. Luego, puede desplegarse en un entorno cloud para lograr una aplicación segura y escalable.

 

8. Desarrollo del interfaz e integración de la aplicación

Si el caso de uso lo requiere, desarrollar una interfaz de usuario (UI) amigable y funcional para que los usuarios interactúen con el asistente. Puede ser una aplicación web, móvil o integrada en otro software. Luego debemos realizar la integración necesaria con las APIs de backend para que cobre vida.

 

9. Monitoreo y alertas

Las herramientas de monitoreo y logging rastrean las interacciones del usuario y las respuestas del asistente. 

Se definen métricas clave de rendimiento (KPIs) y establecen umbrales para alertas. Por ejemplo, si el asistente no puede responder a un cierto porcentaje de preguntas.

 

10. Iteración y mejora continua

Los datos recolectados a través del monitoreo permiten identificar áreas de mejora. También se podría solicitar retroalimentación a los usuarios y a partir de este feedback realizar ajustes. 

Esto puede incluir ajustar el prompt, afinar el embedding o expandir la base de datos con nuevas fuentes de información.

Estos pasos dan una estructura sólida para crear, implementar y mantener un asistente conversacional impulsado por IA.

 

 11. Conclusiones

– El primer paso antes de construir un asistente conversacional basado en IA es entender qué valor aportará al negocio.

– Con eso resuelto, se avanzará sobre los requerimientos y los objetivos puntuales del proyecto y con la identificación de las fuentes de datos que le darán contexto al asistente.

– Luego se construyen embeddings para interpretar el contenido de los documentos y bases de datos vectoriales para buscar respuestas o sugerencias basadas en similitud semántica.

– El siguiente paso es esencial: diseñar los prompts que guíen al modelo a generar respuestas alineadas con el propósito. Hecho esto, se utilizan API para que el modelo esté disponible para los usuarios.

– El desarrollo de una interfaz amigable es la clave para una mayor adopción. No obstante, gracias al monitoreo y el logging es posible verificar si el funcionamiento del asistente se da de acuerdo a lo esperado.

– Y este es apenas el principio del recorrido: siempre existen oportunidades para mejorar el asistente.

 

Nos gustaría acompañarte en el desarrollo de tus asistentes conversacionales basados en IA para que potencies las interacciones con tus clientes.

 

¡Agenda tu reunión con un especialista!

Analía Laura Enrique

About Analía Laura Enrique