📍 Dublin, Ireland

I am a Software and AI Engineer with an MSc in Data Analytics, a BSc (Hons) in Information Technology, and an Industrial Informatics Technician qualification.

My interest in technology started with an 8-bit MSX and cassette tape, then grew through industrial informatics, electronics, digital logic, microprocessors, automation, technical support, software development, cloud systems, data engineering, and machine learning.

Today, I am particularly interested in the point where AI meets real engineering: efficient inference, model optimization, edge and embedded AI, reproducible ML pipelines, distributed systems, observability, and production-oriented infrastructure.

An independent, reproducible CPU inference study comparing PyTorch FP32, OpenVINO FP32, and NNCF INT8 post-training quantization using a pretrained ResNet18 model.

On the measured workload, INT8 reduced the OpenVINO model footprint by 3.94× and produced a 7.46× mean-latency ratio improvement versus OpenVINO FP32, while changing Imagenette Top-1 accuracy by only -0.076 percentage points.

The project includes:

• PyTorch to OpenVINO model conversion
• NNCF post-training quantization
• real-image calibration
• Imagenette validation
• latency and throughput benchmarking
• numerical consistency checks
• reproducible reports
• automated chart generation
• regression tests
• environment capture

Stack: Python · PyTorch · torchvision · OpenVINO · NNCF · NumPy · pandas · matplotlib · pytest

→ View project

Portfolio version of my MSc Data Analytics dissertation, investigating progression-free survival prediction in soft-tissue sarcoma using clinical and radiomic features.

The project implements an end-to-end medical imaging and survival machine learning pipeline covering:

• DICOM and RTSTRUCT mapping
• NIfTI conversion
• tumour segmentation masks
• PyRadiomics feature extraction
• patient-level feature aggregation
• clinical and radiomic data integration
• leakage-aware preprocessing
• Cox Proportional Hazards modelling
• DeepSurv neural survival modelling
• bootstrap evaluation
• Kaplan-Meier risk stratification

Stack: Python · PyTorch · PyRadiomics · CoxPH · DeepSurv · scikit-learn · pandas · NumPy · DICOM · NIfTI

→ View project

An event-driven engineering reference platform for deterministic incident intake and asynchronous, retrieval-grounded AI analysis.

The platform combines:

• FastAPI
• PostgreSQL and pgvector
• transactional outbox pattern
• Apache Kafka
• local Ollama inference
• retrieval-augmented generation
• controlled AI tool execution
• explicit human review
• OpenTelemetry
• Prometheus
• Tempo
• Grafana
• Docker
• Kubernetes
• Helm

The architecture separates application, messaging, AI, retrieval, persistence, and observability concerns while preserving durable execution and auditability.

Stack: Python · FastAPI · PostgreSQL · pgvector · Apache Kafka · Ollama · OpenTelemetry · Prometheus · Tempo · Grafana · Docker · Kubernetes · Helm

→ View project

Selected contributions accepted and merged into upstream open-source projects.

• #1849: Route assignment trace diagnostics to stderr

Fixed assignment trace diagnostics so linker trace output is consistently routed to stderr, with regression coverage for assignment and FILL/padding traces.

• #22113: Extend CMSIS-Pack smoke test to Cortex-M

Extended the CMSIS-Pack smoke test to exercise the Cortex-M and CMSIS-NN path on ARMCM55, including a Cortex-M quantized operator and CMSIS-NN integration.

• #304: Fix mixed value type decoding in sds-convert

Fixed mixed-width binary data decoding where channel-by-channel parsing could misalign samples containing different numeric value sizes.

Rust · C++ · Python · Node.js · Java

ExecuTorch · PyTorch · TensorFlow · OpenVINO · NNCF · scikit-learn · Apache Spark · pandas · NumPy · PyRadiomics · Deep Learning · Survival Analysis · Model Optimization · Quantization

FastAPI · Flask · REST APIs · PostgreSQL · pgvector · Apache Kafka · SQL · NoSQL

Linux · Docker · Kubernetes · Helm · Jenkins · GitHub Actions · CI/CD · Cloud Architecture

OpenTelemetry · Prometheus · Grafana · Tempo · Distributed Tracing · Structured Logging

AWS · Azure · Google Cloud · Firebase

Debugging · Root-Cause Analysis · Reverse Engineering · Systems Integration · Performance Analysis · Regression Testing · Reproducible Experimentation

I am especially interested in engineering problems involving:

• AI inference optimization
• model compression and quantization
• edge and embedded AI
• ML runtime systems
• PyTorch, OpenVINO and ExecuTorch ecosystems
• C++ inference infrastructure
• AI platforms
• distributed backend systems
• Kubernetes-based ML infrastructure
• performance engineering
• observability
• production machine learning systems

My journey into technology began long before AI became mainstream.

I started experimenting with computers on an 8-bit MSX using cassette tape storage. During my qualification as an Industrial Informatics Technician, I studied electronics, digital logic, microprocessors, computer architecture, industrial computing, and automation.

That foundation led to years of hands-on work with computers, systems troubleshooting, technical support, and software technologies, followed by formal higher education in Information Technology and Data Analytics in Ireland.

Today I am combining that background with modern AI and software engineering, contributing to open-source projects involving model optimization, inference runtimes, embedded AI, compilers, linkers, and low-level systems software.

CCT College Dublin 2025

Main areas:

Machine Learning · Deep Learning · Data Mining · Data Visualization · Statistics · Cloud Computing · Research Methods · Applied Data Analytics

Dissertation focus:

Radiomics-based survival prediction in soft-tissue sarcoma using CoxPH and DeepSurv

CCT College Dublin 2023

Main areas:

Python · Java · Machine Learning · AI · Data Mining · SQL · NoSQL · Node.js · Cloud Architecture · Software Engineering · Strategic IT

ETE João Baptista de Lima e Figueiredo Mococa, São Paulo, Brazil 1996

Technical training in:

Digital Electronics · Digital Logic · Microprocessors · Computer Architecture · Industrial Computing · Automation

I am building toward engineering roles where machine learning meets systems engineering.

That includes:

Software Engineering · AI Engineering · ML Systems · Model Optimization · Edge AI · Embedded AI · ML Platform Engineering · AI Infrastructure

I particularly enjoy engineering work that requires understanding a system deeply, reproducing a failure, identifying the root cause, validating the fix, and leaving behind regression coverage so the same problem does not return.

LinkedIn linkedin.com/in/ricardo-alves-de-souza

GitHub github.com/ricardoasouz

Location Dublin, Ireland