3 days ago
6
Irshad Buchh is a seasoned technologist with over 30 years of experience in the tech industry, currently working as a Cloud Solutions Engineer at Oracle, where he deploys large-scale AI/ML and GPU clusters to train and build Large Language Models for various use cases, focusing on industries such as healthcare, startups, and manufacturing. Before this, he served as a Principal Solutions Architect at AWS from April 2019 to June 2024, playing a key role in cloud implementations across industries, including healthcare. Recognized for his thought leadership and mentorship, Irshad has been a guiding force for numerous engineers and architects in the field of cloud computing.
With over 20 publications listed in his Google Scholar profile and a distinguished track record of speaking at prominent industry conferences such as various IEEE events, AWS re: Invent, Oracle Cloud World, CdCon, DockerCon, and KubeCon, Irshad is a leading figure in the cloud and AI/ML fields.
In this exclusive thought-leader article with AI Time Journal, we have the privilege of speaking with Irshad Buchh about his innovations in Cloud and AI/ML.
With over 30 years of experience in technology, including 9 years in cloud computing and 5 years in AI/ML, how has your career trajectory shaped your approach to building machine learning models and AI-powered generative applications?
My career journey has been an evolution of learning and adapting to emerging technologies. Starting with traditional software engineering and systems design, I developed a strong foundation in problem-solving and a deep understanding of system architectures. As the industry shifted towards the cloud, I embraced this change early, focusing on how cloud platforms could enable scalability, flexibility, and innovation.
Over the past nine years in cloud computing, I’ve worked extensively with leading platforms like AWS and Oracle Cloud, helping organizations migrate, modernize, and optimize their workloads. This experience gave me a unique perspective on how cloud infrastructure could accelerate AI/ML workflows. When I transitioned into AI/ML about five years ago, I realized the transformative potential of combining these technologies.
In building machine learning models and generative AI applications, I approach projects with a blend of engineering rigor and a keen eye on business outcomes. My goal is to design solutions that are not just technically robust but also aligned with user needs and scalability requirements. For instance, while working with startups, I’ve seen how cloud-based generative AI can empower businesses to create innovative products, often overcoming resource constraints.
Additionally, my mentorship role with data scientists has taught me the importance of collaboration and knowledge sharing in the rapidly evolving AI landscape. These experiences have shaped my belief that the best AI solutions are born at the intersection of cutting-edge technology and practical, user-focused application.
Can you discuss the role of cloud platforms, particularly Oracle Cloud, in democratizing access to AI and machine learning for startups and enterprises?
Oracle Cloud plays a pivotal role in democratizing access to AI and machine learning, particularly through its advanced GPU clusters and Kubernetes support. The availability of NVIDIA GPUs like A100 and H100 on Oracle Cloud provides immense computational power necessary for training complex machine learning models, including large language models (LLMs) and generative AI applications. These GPU clusters are designed to handle data-intensive workloads, offering high performance and scalability at a fraction of the cost compared to on-premises solutions.
Using Oracle Kubernetes Engine (OKE) in tandem with GPU clusters further enhances the flexibility and efficiency of building and deploying ML models. Kubernetes simplifies the orchestration of containerized workloads, allowing teams to scale training jobs dynamically based on demand. This capability is particularly valuable for startups and enterprises looking to optimize resource utilization and cost efficiency.
For instance, with OKE, you can deploy machine learning pipelines that automate data preprocessing, model training, and hyperparameter tuning. The integration of Kubernetes with Oracle’s GPU clusters enables distributed training, which significantly reduces the time required for model development. This combination also supports the deployment of inference services, making it seamless to integrate trained models into production systems.
Startups often leverage this setup to experiment with state-of-the-art AI capabilities without the need for extensive infrastructure investment. Similarly, enterprises use Oracle’s GPU-enabled Kubernetes solutions to modernize their workflows, enabling AI-driven automation, enhanced analytics, and real-time decision-making.
In my experience, this synergy between GPU clusters and Kubernetes on Oracle Cloud has been a game-changer, allowing teams to focus on innovation while the platform handles scalability, reliability, and performance optimization. This truly embodies the democratization of AI and ML, making these technologies accessible to a broader audience, irrespective of their size or budget.
Generative AI has gained significant traction recently. What are some of the most exciting real-world applications you have been involved with, and what challenges did you face during their implementation?
One of the most impactful generative AI applications I have architected is a solution designed specifically for medical professionals to streamline the creation of clinical notes during patient visits. This application, deployed on laptops and iPads, leverages generative AI to record doctor-patient conversations (with the patient’s consent) and automatically generate comprehensive clinical notes based on the dialogue.
The workflow is intuitive: as the conversation unfolds, the application not only transcribes the dialogue but also integrates medical terminology and IC9 codes from a vast medical knowledge base. After the visit, the doctor can review, make necessary edits, and approve the clinical notes, which are then seamlessly saved into the Electronic Health Record (EHR) system.
This system has been transformative for both patients and doctors. Patients appreciate the enhanced face-to-face interaction, as doctors no longer need to divert their attention to manual note-taking. For doctors, the solution significantly reduces the administrative burden, freeing up time to focus on patient care while ensuring accurate and complete documentation.
Challenges and Solutions:
- Data Privacy and Consent:
Recording sensitive conversations in a clinical setting raised concerns about data security and patient privacy. To address this, we implemented robust encryption protocols and secured patient consent workflows to ensure compliance with HIPAA and other data privacy regulations.
- Medical Knowledge Base Integration:
Incorporating IC9 codes and ensuring the accuracy of medical terminology required extensive collaboration with domain experts and the use of a comprehensive, continually updated medical knowledge base.
- Real-Time Processing:
Ensuring that the transcription and generation of clinical notes occurred in real-time without compromising the system’s responsiveness was another challenge. We optimized the application by leveraging Oracle’s GPU-powered cloud infrastructure, which facilitated efficient processing and inference.
- User Adoption and Training:
Convincing doctors to trust and adopt the system required addressing their concerns about accuracy and ease of use. We conducted extensive user testing, provided training sessions, and incorporated feedback to refine the interface and improve reliability.
This project demonstrated the transformative potential of generative AI in the healthcare sector, making routine tasks less burdensome and enhancing the overall experience for both patients and doctors. It was incredibly rewarding to see how technology could make such a meaningful impact on people’s lives.
Your recent paper, ‘Enhancing ICD-9 Code Prediction with BERT: A Multi-Label Classification Approach Using MIMIC-III Clinical Data,’ published in IEEE, explores an intriguing application of AI in healthcare. Can you elaborate on the key findings of this research and its potential implications for improving healthcare practices?
In my recent research paper, I addressed the critical challenge of automating ICD-9 code assignment from clinical notes, focusing specifically on ICU records in the MIMIC-III dataset. Leveraging the power of BERT, we demonstrated how transformer-based models can significantly improve prediction accuracy over traditional methods like CAML, which primarily rely on convolutional neural networks.
One of the key innovations in our study was the preprocessing pipeline to handle BERT’s sequence length constraints. By implementing automatic truncation and section-based filtering, we optimized input data to fit the model while preserving essential clinical information. This allowed us to fine-tune BERT effectively on the top 50 ICD-9 codes, achieving a competitive Micro-F1 score of 0.83 after just one epoch using 128-length sequences.
The potential implications of this work are substantial. Automating ICD-9 code assignments with high accuracy can greatly reduce the manual workload for healthcare professionals and ensure consistent coding practices. This, in turn, improves patient data management and facilitates better healthcare analytics. Future efforts will focus on extending sequence lengths and comparing performance with other preprocessing methods to further refine the approach.
By demonstrating the potential of transformer-based architectures like BERT in healthcare informatics, this research paper provides a robust framework for developing scalable and efficient solutions that can transform clinical workflows and enhance the overall quality of care.
With the growing adoption of AI and cloud-based NLP solutions among small and medium-sized businesses, what challenges and opportunities do you foresee for enterprises in leveraging these technologies for predictive market analysis and consumer intelligence? How do cloud-based tools contribute to addressing these needs?
The growing adoption of AI and cloud-based Natural Language Processing (NLP) solutions among small and medium-sized businesses (SMBs) represents a transformative shift in how organizations approach predictive market analysis and consumer intelligence. However, this shift brings both opportunities and challenges.
Opportunities: Cloud-based NLP solutions democratize access to advanced AI capabilities, enabling SMBs to compete with larger enterprises. These tools allow businesses to process vast amounts of unstructured data—such as customer reviews, social media interactions, and feedback—at scale. For instance, AI-powered chatbots and voice-enabled NLP systems can provide real-time insights, helping SMBs optimize customer experience (CX) and make informed decisions about market trends.
Challenges: The primary challenge is managing the complexity of implementation and integration into existing workflows. SMBs often lack technical expertise and resources, which can hinder the adoption of these solutions. Additionally, data privacy and compliance with regulations like GDPR and CCPA are critical, particularly when handling sensitive consumer data. Scalability can also be an issue, as businesses must balance the costs of processing increasing volumes of data with their operational budgets.
How Cloud-Based Tools Help: Cloud platforms like Oracle Cloud provide scalable, secure, and cost-effective solutions tailored to SMBs’ needs. For example, Oracle’s AI/ML offerings simplify the deployment of NLP applications through pre-built APIs and no-code/low-code tools. These solutions enable businesses to extract actionable insights without the need for extensive technical expertise.
Moreover, Oracle’s GPU-accelerated clusters and robust data integration capabilities support complex workloads such as predictive modeling and real-time analytics. These tools empower SMBs to not only harness the power of NLP but also adapt quickly to changing consumer demands. By lowering barriers to entry and offering secure, scalable infrastructure, cloud-based tools ensure that SMBs can fully leverage AI and NLP to drive innovation and growth in a competitive market.
How do you see advancements in NLP technologies, particularly in auto-coding and text analytics, shaping industries such as compliance, risk management, and threat detection? Can you elaborate on how these technologies uncover hidden patterns and anomalies, and share any relevant experiences from your work in deploying such solutions in the cloud?
Advancements in NLP technologies, particularly in auto-coding and text analytics, are revolutionizing industries like compliance, risk management, and threat detection by enabling a deeper, faster, and more accurate analysis of structured and unstructured data. Auto-coding, for example, automates the tagging of data with relevant categories, making it easier for compliance teams to identify critical information and anomalies. This is achieved using techniques such as topic modeling, sentiment analysis, and clustering, which extract meaningful patterns from large datasets.
At Oracle, we leverage cloud-based NLP solutions to process and analyze massive volumes of data efficiently. For instance, in compliance scenarios, NLP models deployed on Oracle’s high-performance GPU clusters are used to scan financial transactions or communication logs for indicators of fraudulent activity or policy violations. The use of techniques like Named Entity Recognition (NER) allows these models to identify key entities and relationships within text, while sentiment analysis can flag negative sentiment that may indicate risks.
In threat detection, NLP-powered tools are instrumental in processing data from diverse sources, including social media and customer feedback, to uncover potential security threats. These tools rely on pattern recognition algorithms to detect anomalies and deviations from expected behaviors. Oracle’s scalable cloud infrastructure ensures that these models can process data in near real-time, providing organizations with actionable insights for preemptive measures.
Our work aligns closely with these advancements as we continually optimize NLP pipelines for accuracy and efficiency. For example, we use Oracle Cloud’s managed Kubernetes clusters to orchestrate and deploy microservices for data preprocessing, model inference, and reporting. These services seamlessly integrate with Oracle Autonomous Database for secure storage and retrieval of insights, providing a robust and scalable solution tailored to the demands of modern enterprises.
Given your mentoring experience with data scientists transitioning to cloud-based workflows, what advice would you give to professionals looking to excel in building AI and generative AI applications in the cloud?
Mentoring data scientists transitioning to cloud-based workflows has been one of the most rewarding aspects of my career. For professionals looking to excel in building AI and generative AI applications in the cloud, my advice centers around three key pillars: learning, adaptability, and collaboration.
- Deepen Your Technical Foundations:
A strong understanding of core cloud services—computing, storage, networking, and databases—is essential. Familiarize yourself with cloud platforms like Oracle Cloud, AWS, and others. Learn about specific services for AI workloads, such as GPU instances, Kubernetes for orchestration, and storage solutions optimized for large datasets. Mastering tools like Terraform for automation or Python for development will also greatly enhance your capabilities.
- Embrace Specialized AI Workflows:
Generative AI applications often require specific infrastructure, like high-performance GPUs for training models and scalable compute for inference. Get comfortable working with ML frameworks like TensorFlow, PyTorch, or Hugging Face for fine-tuning generative models. Understanding data preprocessing pipelines and model deployment strategies, such as containerized deployments on Kubernetes clusters, will set you apart.
- Collaborate Across Disciplines:
Generative AI projects often involve cross-functional teams, including data scientists, cloud engineers, domain experts, and business stakeholders. Effective communication and collaboration are crucial. Be proactive in understanding the goals and constraints of all stakeholders and ensure alignment throughout the project lifecycle.
- Stay Current and Experiment:
AI and cloud technologies are evolving rapidly. Stay up to date with advancements like fine-tuning large language models, leveraging pre-built APIs, or adopting hybrid cloud strategies. Experimenting with open-source projects and participating in hackathons can help you explore new ideas and build a strong portfolio.
What advancements or trends in AI/ML and cloud computing do you see shaping the next decade, and how are you preparing to lead in this rapidly evolving space?
The next decade promises to be transformative for AI/ML and cloud computing, with several key advancements and trends expected to shape the landscape. As someone deeply immersed in both fields, I see the following trends as particularly impactful:
- Rise of Generative AI and Large Language Models (LLMs):
The rapid evolution of generative AI, particularly large language models (LLMs) like GPT-4 and beyond, will continue to revolutionize industries such as healthcare, finance, education, and entertainment. These models will not only be used for content creation but also in complex applications such as personalized medicine, autonomous systems, and real-time decision-making. In my work, I’m preparing for this shift by focusing on the integration of LLMs with domain-specific knowledge, leveraging cloud platforms to make these powerful models accessible and scalable for businesses of all sizes.
- AI-Powered Automation and MLOps:
As businesses scale their AI initiatives, automation will become crucial. MLOps—the practice of applying DevOps principles to machine learning—will enable companies to streamline their AI workflows, from model development to deployment and monitoring. This trend will democratize AI by making it more efficient and accessible. I am preparing for this by gaining deeper expertise in cloud-based AI tools like Kubernetes for orchestrating machine learning models and leveraging services like Oracle Cloud’s GPU clusters to accelerate AI workloads. These advancements will enable organizations to focus more on innovation while leaving the heavy lifting to automated systems.
- Edge Computing and AI at the Edge:
The shift to edge computing is gaining momentum, where data processing happens closer to the source of data generation (e.g., IoT devices, mobile devices). This allows for real-time decision-making and reduces the latency associated with cloud-based processing. With advancements in 5G and IoT, edge AI will become even more prevalent, especially in industries such as healthcare (e.g., wearable devices), autonomous vehicles, and smart cities. I’m actively involved in developing cloud-based solutions that integrate edge AI, ensuring that the infrastructure I architect supports both cloud and edge computing models seamlessly.
To lead in this rapidly evolving space, I am focusing on continuous learning and staying ahead of these trends. I am deeply involved in the cloud and AI communities, contributing to thought leadership through articles and speaking engagements, while also working on practical, real-world applications of these technologies. Additionally, I mentor emerging AI professionals and collaborate with cross-functional teams to drive innovation. By maintaining a forward-looking mindset and embracing the power of cloud computing, I am well-positioned to help organizations navigate this exciting future.
English (US) ·