Turning Data Into Decisions: Saddam Hussain on AI, Data Engineering, MLOps, and Intelligent Industry

My journey began with Electrical and Electronics Engineering, which gave me a strong foundation in systems, automation, and engineering processes. Over time, I became increasingly interested in the data generated by those systems and how it could be used to improve decisions and solve practical problems.

That interest gradually led me into analytics, data engineering, machine learning, and artificial intelligence. I later pursued graduate studies in Business Analytics and Decision Analytics and am currently pursuing a Master of Science in Artificial Intelligence.

My professional experience has taken me across manufacturing, automotive, healthcare, and industrial operations. While the industries have been different, the underlying question has remained consistent: What is the real problem, and how can data and technology be used to solve it?

At Carmeuse Lime & Stone, my work has involved industrial data pipelines, machine-learning models, operational dashboards, KPI development, time-series analytics, automated reporting, and model deployment. My earlier experience has also included EV battery health management, vehicle data, healthcare analytics, machine-failure prediction, aerospace defect classification, enterprise data integration, and vehicle-tracking systems.

This combination has shaped my view that data science cannot operate in isolation. The strongest solutions bring together domain knowledge, data engineering, analytics, machine learning, deployment, and an understanding of the people who ultimately use the system.

I see data engineering as one of the foundations of successful AI. Before building a model, we need to know whether the data is complete, timely, consistent, and trustworthy.

In real industrial organizations, information rarely comes from one clean source. It can come from SQL databases, SAP, process historians, sensors, APIs, InfluxDB, Excel files, equipment systems, and streaming platforms. These sources have to be connected, validated, aligned, and transformed before they can reliably support analytics or machine learning.

At Carmeuse, I have worked with data from SQL, SAP, InfluxDB, process historians, and other plant systems. Data engineering makes that information usable, while data science applies it to machine learning, optimization, anomaly detection, forecasting, and analysis.

One example involved automated quality-data collection. The pipeline reduced manual errors by approximately 95% and improved operator efficiency by around 20%. These improvements demonstrate why the data layer matters. If the underlying information is unreliable, even a technically sophisticated model can produce unreliable outcomes.

I see data engineering and data science as parts of the same solution rather than separate disciplines. The quality of an AI system ultimately depends on the quality and reliability of the information supporting it.

My work with General Motors on Vehicle Health Management and EV battery data gave me direct experience with the challenges of operating at significant data scale. The platform processed approximately 35 million telemetry messages per day and supported around 8,000 customers daily.

At that scale, data engineering becomes considerably more complex. Messages can arrive late or out of order, but they still need to be associated with the correct vehicle and event time. The system also has to process information reliably enough to support diagnostics and AI-driven alerts.

We worked with Databricks, Azure Event Hubs, Spark Structured Streaming, ETL pipelines, and Medallion architecture. The work contributed to a 30–40% reduction in EV battery downtime and supported battery diagnostics and AI-driven alerts. Migrating the Vehicle Health Management platform to Databricks also reduced code complexity by approximately 60%, lowered infrastructure costs by around 20%, and improved algorithm performance by approximately 30%.

Medallion architecture was particularly useful because it creates a structured progression from raw information to trusted, analytics-ready information. Raw data can be preserved while subsequent layers clean, validate, and prepare it for downstream use.

That experience later became the foundation for my research paper, “Medallion-Based Data Engineering for EV Battery Health Monitoring: Managing 35M Daily Telemetry Messages at Scale.”

The biggest lesson I took from that work is that scalable AI begins long before a model is trained. It begins with how data is captured, processed, validated, governed, and made available.

Industrial environments are particularly interesting because many variables interact simultaneously. In lime production, for example, factors such as fuel conditions, airflow, temperatures, oxygen levels, kiln speed, feed conditions, and material properties can all influence quality and process performance.

At Carmeuse, I have worked on machine-learning models around quality measures such as residual CO₂ and sulfur. The objective is to understand the relationships between process conditions and quality outcomes and identify operating conditions associated with better performance.

Some of this work contributed to approximately a 3-4% improvement in lime quality.

What makes industrial AI valuable is that improvements are often interconnected. A change in one area can affect quality, production, process stability, energy consumption, emissions, and operational decision-making.

I have also developed centralized operational dashboards in Grafana supporting more than 10 U.S. plants and two Canadian plants. These dashboards brought production, quality, energy, equipment performance, and operational KPIs into a common environment. The automation reduced close to a week of manual reporting effort each month, while standardized KPIs helped teams identify deviations earlier and contributed to approximately a 5% reduction in plant downtime.

For me, the objective of industrial AI is not simply to create another model. It is to give teams better information and decision support that can translate into measurable operational improvements.

AI has applications beyond production efficiency, and workplace safety is an important example.

I developed a computer-vision solution for Personal Protective Equipment compliance in heavy-machinery environments. The system supported safety monitoring across multiple locations and contributed to a reported 70% reduction in safety violations.

I also developed a facial-recognition-based access-control solution for a critical server-room environment. The system was designed to identify authorized users, detect unauthorized attempts, and record associated images and event metadata.

These projects showed me how computer vision can be applied to practical operational challenges. Instead of treating AI as an abstract technology, we can connect it to specific safety requirements and organizational processes.

This work also contributed to my invited talk at TIC 2026, titled “Computer Vision Applications for Industrial Safety and Smart Workplace Monitoring,” where I discussed the use of AI in supporting zero-injury workplaces. I also served as a Session Chair.

The important consideration with such systems is that technology should support well-defined safety processes and organizational objectives. AI can provide monitoring and alerts, but successful implementation also depends on how those insights are incorporated into operational workflows.

Building a machine-learning model is only one part of the journey. Once a model moves into production, deployment, monitoring, versioning, governance, maintenance, and reuse become equally important.

I have worked on automated CI/CD pipelines for machine-learning applications that reduced deployment time by approximately 75% and supported model reuse across multiple locations.

This becomes particularly important in industrial environments because different plants can have different equipment, infrastructure, operating conditions, and data patterns. If every plant requires an entirely new implementation, scaling AI becomes expensive and difficult to maintain.

My work in this area became the foundation for the paper “End-to-End MLOps for Multi-Plant Industrial AI: Deployment Automation, Model Reuse, and Governance,” which I presented at EAMCON 2026.

The broader lesson is that organizations need repeatable processes for deploying and maintaining AI. Monitoring, validation, configuration management, governance, and version control are all part of making AI reliable in production.

I am particularly interested in how successful solutions can be reused across locations without rebuilding the entire system each time. MLOps provides an important framework for making that possible.

My research is closely connected to the problems I encounter in industry. I believe practical challenges can provide valuable research questions, while research can help create structured approaches to solving those challenges.

My EV battery data-engineering research is one example. The paper, “Medallion-Based Data Engineering for EV Battery Health Monitoring: Managing 35M Daily Telemetry Messages at Scale,” examined challenges around streaming data, late and out-of-order messages, data quality, scalable architecture, and trusted data layers. I presented the work at an IEEE conference at MAHSA University in Kuala Lumpur in 2026, where it received the Best Paper Award.

At ICNCDA 2026, I also delivered a keynote on optimizing industrial quality and emissions simultaneously through machine learning. My ongoing research includes time-series analysis and residual CO₂ prediction in rotary-kiln operations.

Another important part of my academic contribution is peer review. During 2026, I reviewed more than 50 research papers for venues including ACIS, Arabian Journal for Science and Engineering, ARIIA, GICITE, ICIEEE, ICSPCRE, and WISE. The research areas included machine learning, computer vision, healthcare, biomedical applications, data science, AI, and AI agents.

Reviewing has made me more critical of technical work, including my own. It encourages me to ask fundamental questions: Is the problem meaningful? Does the methodology support the claims? Are the conclusions supported by the evidence?

Those questions are equally important in industry before a model is deployed into a production environment.

I would encourage people entering AI and data engineering not to learn these areas in isolation. They should understand databases, ETL, APIs, streaming, cloud platforms, data quality, machine learning, visualization, and MLOps.

Most importantly, understand the problem before choosing the technology.

It is easy to say, “I built a model.” A much stronger explanation is: Why was the model built? Who used it? What decision did it support? What changed because of it?

I would also encourage people to develop both technical depth and domain understanding. AI becomes much more useful when practitioners understand the environment in which their systems operate.

Looking ahead, I want to continue working at the intersection of data engineering, industrial AI, computer vision, time-series analytics, MLOps, and intelligent decision systems. I am particularly interested in systems that help teams understand what may happen next and what action may be useful.

I also plan to continue contributing through research, publications, speaking engagements, and peer review.

For me, the future of AI is not simply about increasingly sophisticated models. It is about building trustworthy systems that turn complex data into useful intelligence and create improvements that people can see, measure, and use. The real opportunity lies in connecting strong data foundations with intelligent systems that solve problems in the real world.