Humberto Ribeiro de Souza

I lead the design, deployment, and adoption of production-grade Generative AI and Agentic AI solutions that deliver measurable business impact, while strengthening core data and platform foundations. Key contributions include: Led Axper’s first end-to-end Generative AI implementation, introducing AI capabilities to the Software Engineering team and collaborating across product, data, and leadership groups Designed and deployed the first Azure AI Foundry–based GenAI architecture, enabling intelligent report summarization and AI-assisted analytics in production Built Axper’s first AI agent capable of generating complete dashboards from natural-language prompts, including persistence, sharing, and reuse — significantly reducing delivery time for customer insights Delivered customer-facing GenAI features that unlocked new customers and generated ROI within one month of launch Evaluated, tested, and classified multiple foundation models, IDEs, and GenAI workflows, defining internal usage guidelines and pre-implementation strategies across cloud providers Trained and onboarded internal engineering teams on practical, safe, and scalable GenAI adoption Streamlined and resolved critical and complex backlogs (fiscal calendars, business hours, extended business logic) while simultaneously delivering new AI-driven capabilities Implemented AI-optimized data flows and data lake improvements to support scalable analytics and agent-driven dashboards Continued ownership of scalable backend systems, data architectures, and robust ETL/ELT pipelines, ensuring data quality, governance, and performance

I led a team effort to enhance the performance of the Train Shipment Visibility (TSV) system, resulting in significant optimizations that directly impacted user experience and operational efficiency. Key achievements include: - Speed Optimization: Reduced end-to-end processing time for the Override Operation by 50%. - Optimizations for side-by-side/compare operations achieved up to a 75% performance gain, reducing full 100+ railcar train load times from 2.1 seconds to just 480 milliseconds. - Data Efficiency: Achieved an 83% reduction in railcars data payload size, bringing down less than a KB. This optimization has ensured consistent UI performance. - UI Responsiveness: Improved refresh rates for 100+ railcar flags, reducing update time to sub-second levels, even on slower devices. - Increased Business Transactions: Technically led the integration between SAP Ariba and ERP Business Central for major global mining clients, enabling over $9 million in transactions in the first year through streamlined, cost-effective AWS-managed services and Python-based AWS Chalice projects. - Enhanced Analytics & Automation: Optimized logistics AI processes through advanced vehicle identification, classification, and data transformation. Leveraged Snowflake, Postgres, ElasticSearch, DynamoDB, MS SQL Server, and both AWS and Azure tools to automate reporting and streamline analytics, employing SQL, views, triggers, and a mix of flat, star, and hybrid schemas. - Corporate System Integration & LUKE AI: Developed backend integrations for the LUKE AI engine using REST APIs, webhooks, streaming, and batching, ensuring seamless connections across corporate systems. - Data-Driven Business Insights: Conducted end-to-end reviews of business processes, enabling intelligent reporting and decision-making via AWS, Azure, and Power BI infrastructures.

Focus on Cloud Systems Integration, continuous improvement, Team Coordination. Coordinated super users and vendors in corporate Systems' improvement planning, support, training development and delivery. Integration design and provisioning of reporting and analytics resources for Finance, HR, Maintenance departments for the ERP and operations supporting systems.

Integrated the current ERP to SAP Concur and finance systems Managed training and administration procedures related to the ERP system, especially coordinating it with Centerra and Hitachi-ABB PowerGrids. Standardized procedures and training materials. Reimplemented the internal and GitHub and cleaned vendors' GitHub related to custom ERP code, integration scripts and system reports Specified API functions, API integration to new departments or acquired companies. Managed data transformation between systems and create custom Jasper Reports and Power BI dashboards both on-site and on Azure cloud, including serverless functions on Azure Functions, Power Automate, SharePoint and others. Support coordination and management activities according to the directors and VPs expectations.

Executed the standardization and technical documentation of the code repository of active projects resulting in a higher quality of delivered code and reducing the number of contractors from 4 to 2 for the same workload. Reduced the cloud costs by 30% by coordinating the transfer of servers to a private cloud solution. Integrated, customized and built custom code for SugarCRM front and backend, mostly using PHP, Javascript, Jenkins, Java, MySQL, Oracle, MS SQL Server and bash running under CentOS Linux on-site, for Metro Inc. Created software development patterns and documented the ingestion pipeline with the use of custom APIs, Pentaho, and other ETL tools between top ERPs and CRM solutions from Microsoft, Oracle, Salesforce & SugarCRM for top Canadian and American Companies. Projects: - Dagostino Electronic Services - www.descomm.com – prepared the Linux CentOS Infrastructure preparation and version migration; - Internal - organized, centralized and monitored the Git repositories; Coordinated the pull requests from remote teams and vendors with AWS EC2, RDS, Route53, VPC, and on-premises servers; - Metro Inc. - updated the Continuous Integration, Continuous Development processes using Jenkins, Java, PHP scripts, Unit tests; - Metro Inc. – customized SugarCRM modules, front-end - Javascript, JQuery and backend - PHP, MySQL; Automate package tests and generation; - Metro Inc. – optimized the performance of the virtual servers, CentOS, MySQL, Redis, Apache 2, Java, Jenkins; - Metro Inc. - created and adjusted of API endpoints to and from the ERP SAP; - Jefo - jefo.ca - co-managed overseas and local development teams, pull requests, code review and Customization of Sugar interface and base functions, mostly in PHP and JavaScript; - Jefo - jefo.ca - standardized, ingested and handled ETL from Rest API, MySQL, XLSX and CSV files to SugarCRM via batches using Pentaho; - Government of Prince Edward Island - implemented a technical file update tool;

Improved by 200% the data processing capacity by migrating custom and independent solutions to a single cloud-hosted system for delivering high-level and standardized AI design patterns to customers. Reduced by 70% the time-to-deliver PoCs - Proof of Concepts - by designing and implementing a big data infrastructure for credit systems and text classification in the AWS cloud environment, as well as a development environment based on EC2, EMR, S3, Route53, and RDS; Conducted a number of integrations and PoCs within the customers with out-of-the-box solutions and custom solutions, connecting to customers data and resources mostly through APIs; Tuning and advised on Python, PHP and Java back-end and front-end codes for semi-supervised and unsupervised machine learning models for numeric predictions, text classification and sentiment analysis. Projects: - Somma System - designed and coded of the first stage of big data back-end that parses clients’ data according to IA models via API, written mostly in Python 3.x, Java EE and shell scripting, compiling C libraries when applicable. Hands-on coding and guidance on AWS tools, GCP, blueprints, Linux Servers, API design, frameworks when required. - Somma Dashboard - defined and implemented the first version of the user interface of the system that organizes and render the user information processed by the Somma, written in PHP and Javascript using the Yii 2 framework and Javascript according to Bootstrap 3 standards; - Somma Modules - implemented diverse modules and tools, as defined by the CCO, using open-source solutions as Solr and ElasticSearch. Provided overall and specific guidance to meet the requirements of CxOs. Managed the international team using Agile/Scrum methodologies and support tools; - Internal - Cloud and general software advisor;

Reduced by 60% the time-to-deliver the weekly reports in the QoS - Performance Team by automating the analysis of the network capacity for 4 million customers; Performed the PoC for the usage of AWS resources to improve the capacity of monitoring and data processing of Telecom systems such as PCRF logs and probes using AWS and GCP tools such as Google Big Query; Adapted out-of-the-box software to analyze and predict customer issues which resulted in a significant reduction of ten of complaints to customer service per month, using custom cloud analytics and tools based on Python, Java EE + Spring framework and Perl; Projects: - AWS Implementation - implemented the AWS for Big Data PoC, revolving the Engineering, Operations, and IT teams by creating the needed parsers written in Perl using threads to connect and synchronize with the IDEN, 3G and LTE/4G networks, including their general logs and ASN1. Created an Oracle database in RDS, adjusted schemas and tables, prepared the ETL via Google Big Query, DynamoDB; Connected further systems through APIs saving around 1% of the department O&M and 0.05% of the CAPEX; - Performance Report generation automation - prepared routines to run via CTRL-M, MIT systems, and Excel macros, saving 60% of the time-to-deliver for the weekly report and 30% of the TTD for the daily report - 3G Offloading - used a startup management style to implement and redirect users having supported devices to the LTE 4G network and relieve the 3G network system, using C++ made programs. The project data sources were the Oracle Databases of the engineering department, Oracle of the Marketing Department, telecom device logs of the Operations department; - Roaming shut down studies and implementation - gathered data of a comprehensive set of systems to define which low-usage and high-cost roaming antennae should be shut down by the engineering, using mainly the same resources of the 3G Offloading project;

Co-designed and managed a social network-based TV Rating Service with massive processing bigdata related techniques, in a way that the system became the most promising in the national market, having some of the top 10 TV channels and Top 10 Marketing agencies as customers; Met the requirement to process Twitter firehose for TV comments, comments processing for 95 TV channels and 15000 programs per week. Managed the technical aspect of on premises, SaaS, PaaS, including Amazon AWS, Locaweb, Bluehost and GCP platforms, mostly with Java EE, Spring Framework and Python code, distributed and stand-alone applications running on Windows Hyper-V and Linux environments, using real, para-virtual and virtual machines; Managed Product Recommendation Services for e-commerce’s and social networks, deployed through REST APIs and made sure the SLA levels were achieved; Projects: TTV - Tuilux TV Leads - Classification and identification of engagement groups. TTV - Tuilux TV - Scaled from Air TV channels to Air and Cable. Full coverture and firehose. TTV - Tuilux TV - prototype to scalable production grade solution dand programming design and programming in Java, JSF, MySQL, MongoDB, Python, Perl, Spring, CSS, HTML using AWS managed resources. Tuilux Recommendation - Maintenance and decommissioning. Sicafi - Hardware, firmware calibration. visual processing adjusts, documentation and decommission.

2015-12 - 2016-03: Updated the analytics product developed above and its usability, generating data and analytics to assist coaches and Olympic athletes to improve performance at the 2016 Brazil Olympic Games, and beyond. 2004-04 - 2011-08: Raised more than CAD 1.0 million from top governmental development agencies in Brazil for the conception and creation of the company's analytics product for the Brazilian Olympic Swimming Team which contributed to boosting the performance of the Team Brazil in international competitions. Managing a multidisciplinary team, designed, integrated and launched products, from hardware to the User Interface, by packing firmware and software components. Streamlined the product deployment, coded the API leveraging Java, Spring Framework, JSF, JSP, C++, Delphi, Oracle databases, video and data processing including USB drivers. Projects: - 2016 Velaqua and IOx2 updates - updated the firmware, Delphi and Java software; Minimized the equipment by updating the communications protocol and using Arduino chipset including auxiliary devices drivers; minor UI interface and analytics updates. - Research, partnership, and funding - launched and documented the projects in partnership with top Brazilian research institutions, namely FINEP, FAPESP, and CNPq; - Velaqua and IOx2 products - created the Velaqua and IOX2 devices, containing embedded real-time software to assess high-performance athletes through analytics. This helped head coaches reach to use in training and performance refinement for swimmers including world champions Cesar Cielo, Ana Marcela Cunha, Poliana Okimoto, Matheus Santana; - Long term project management - successfully managed product development for more than five years. Wrote the project handling guidelines to comply with investors, coaches, potential customers, and universities needs. Extreme events were managed with crisis management, waterfall, Agile Scrum, XP and PMBoK.

Implemented the maintenance team of the ISPS Code project by hiring a 30-people team, acting successfully under high SLA standards in three shifts for two years under the guidance/ advice of the Technology Manager; Managed the senior developers and technical staff for documentation, implementation and updating custom firmware and custom backend software in C++, Java, Informix and MySQL, including installation and top industrial software customization for video processing and compression, including API specification and IoT devices customization.

• Founder of the company. Supervised the field teams. Architected and configured servers based on Windows and Linux to small and medium-sized customers. • Implemented and managed partnerships with business groups from other regions of Sao Paulo state to serve larger customers, such as CPFL, CIESP among others, under SLA. • Developed a group of client-server software solutions in different markets for marketing discovery. Sectors were healthcare, transportation, and sports sectors in Delphi 8 and Java. • Left the company to run an IT & D department, focusing on industrial automation products development.

Concepts and applications in the following main subjects - Deep learning - Intelligent Agents & Reinforcement Learning - Machine Learning

Fellow of Sao Paulo Research Foundation (FAPESP) Fellow of National Counsel of Technological and Scientific Development (CNPq) Second best score of the University in the 2001 MEC's Annual National Universities and Students tests - "Provão" in the 80th percentile nationwide.