Gerdau, a major steel manufacturer, implemented an LLM-based assistant to support employee re/upskilling as part of their broader digital transformation initiative. This development came after transitioning to the Databricks Data Intelligence Platform to solve data infrastructure challenges, which enabled them to explore advanced AI applications. The platform consolidation resulted in a 40% cost reduction in data processing and allowed them to onboard 300 new global data users while creating an environment conducive to AI innovation.
Gerdau is Brazil’s largest steel producer with a 123-year history, operating globally with over 30,000 employees. As one of the leading producers of long steel in the Americas and special steel worldwide, the company has embraced digital transformation to optimize manufacturing processes, supply chains, and environmental sustainability initiatives. This case study documents their journey from a fragmented, homegrown open source data ecosystem to a unified data intelligence platform, culminating in their first production deployments of large language model (LLM) applications.
The company’s environmental credentials are notable—they are Latin America’s largest recycling company, with 71% of their steel produced from scrap, and they maintain carbon emissions at roughly half the global industry average. Their digital transformation initiatives directly support these sustainability goals, particularly through their digital twins use case for manufacturing optimization.
Before adopting their new platform, Gerdau’s technical teams had built an impressive but problematic ecosystem of homegrown and open source data tools. While the decision to use open source was driven by flexibility and unlimited potential, the resulting infrastructure created significant operational challenges:
For a steel manufacturer where chemical composition accuracy is critical, having incorrect data at the wrong time could lead to costly production failures. The business impact was clear: these technical hurdles were preventing Gerdau from achieving their strategic goals around ESG commitments, supply chain management, and AI advancement.
Gerdau adopted the Databricks Data Intelligence Platform to consolidate their various tools into a single, user-friendly environment. The implementation involved several key components:
Delta Lake provided the foundational optimized storage layer for the new data infrastructure. This open-source storage format enabled reliable data lakes with ACID transactions, scalable metadata handling, and unified streaming and batch data processing.
Delta Sharing enabled secure data sharing both internally across teams and externally with manufacturing and distribution partners. This was particularly important given Gerdau’s B2B operations involving complex ecosystems of suppliers, distributors, regulators, and customers.
Photon, the next-generation query engine on the Databricks Platform, delivered dramatic performance improvements. Data processing time was reduced from an average of 1.5 hours to just 12 minutes—a significant gain for tables that are processed daily, directly translating to both performance and cost benefits.
Unity Catalog established data governance standards across manufacturing processes, implementing fine-grained access controls, data lineage tracking, and access segregation for different user groups. This also integrated well with Power BI, enabling business teams to create their own reports and dashboards while maintaining governance standards.
While much of the case study focuses on data infrastructure modernization, the GenAI and LLMOps aspects represent the forward-looking portion of Gerdau’s transformation. The unified data platform has enabled the company to explore and deploy cutting-edge AI applications:
LLM-Powered Training Assistant: One of Gerdau’s first achievements using large language models is an AI assistant designed to help employees with re-skilling and upskilling journeys. This represents a production deployment of an LLM application within the enterprise, suggesting the company has established at least the foundational infrastructure for LLMOps workflows including model serving, inference, and user interaction management.
Advanced Analytics and ML Capabilities: Beyond the training assistant, the platform has enabled exploration of:
The case study suggests that having a unified data platform was a prerequisite for these advanced AI initiatives. The improved data governance, lineage tracking, and access controls provided by Unity Catalog would be essential for any production LLM deployment, ensuring that models have access to high-quality, governed data while maintaining compliance requirements.
While the case study does not provide deep technical details on their LLM deployment architecture, several LLMOps-relevant themes emerge:
Data Foundation for LLMs: The emphasis on data unification, governance, and quality directly supports LLM applications. Production LLMs require reliable, well-governed data for both training (or fine-tuning) and retrieval-augmented generation patterns. Gerdau’s investment in Unity Catalog provides the lineage and access control capabilities necessary for responsible LLM deployment.
User Accessibility: The platform reduced the barrier to entry for data work—users no longer needed to learn multiple tools or be Python/Spark proficient. This democratization likely extends to AI/ML workflows, potentially enabling broader organizational participation in LLM application development and use.
Scalability for AI Workloads: The performance improvements (processing time reduced from 1.5 hours to 12 minutes) and cost savings (40% reduction in data processing costs, 80% in streaming development) suggest infrastructure that can handle the computational demands of LLM inference and other AI workloads.
Multi-Regional Deployment: The rapid onboarding of 300 new global data users across operations in Peru, the United States, and Brazil indicates the platform’s ability to support geographically distributed AI applications—an important consideration for enterprise LLM deployments that must serve users across multiple regions.
The quantified results from Gerdau’s transformation include:
These operational improvements have positioned Gerdau for continued AI innovation. The platform’s scalability and flexibility support their growth and expansion strategies, while the governance capabilities ensure compliance with data security standards and regulations.
It’s worth noting that this case study comes from Databricks’ customer stories section, meaning it naturally emphasizes positive outcomes. Some observations to consider:
Nevertheless, Gerdau’s journey illustrates a common pattern in enterprise LLMOps: organizations must first establish solid data foundations, governance frameworks, and unified platforms before they can effectively deploy production LLM applications. The training assistant represents an initial foray into GenAI that builds on this foundation, with the platform positioning them for more ambitious AI applications in predictive maintenance, classification, and other manufacturing use cases.
New Relic, a major observability platform processing 7 petabytes of data daily, implemented GenAI both internally for developer productivity and externally in their product offerings. They achieved a 15% increase in developer productivity through targeted GenAI implementations, while also developing sophisticated AI monitoring capabilities and natural language interfaces for their customers. Their approach balanced cost, accuracy, and performance through a mix of RAG, multi-model routing, and classical ML techniques.
Novartis partnered with AWS Professional Services and Accenture to modernize their drug development infrastructure and integrate AI across clinical trials with the ambitious goal of reducing trial development cycles by at least six months. The initiative involved building a next-generation GXP-compliant data platform on AWS that consolidates fragmented data from multiple domains, implements data mesh architecture with self-service capabilities, and enables AI use cases including protocol generation and an intelligent decision system (digital twin). Early results from the patient safety domain showed 72% query speed improvements, 60% storage cost reduction, and 160+ hours of manual work eliminated. The protocol generation use case achieved 83-87% acceleration in producing compliant protocols, demonstrating significant progress toward their goal of bringing life-saving medicines to patients faster.
Snorkel developed a specialized benchmark dataset for evaluating AI agents in insurance underwriting, leveraging their expert network of Chartered Property and Casualty Underwriters (CPCUs). The benchmark simulates an AI copilot that assists junior underwriters by reasoning over proprietary knowledge, using multiple tools including databases and underwriting guidelines, and engaging in multi-turn conversations. The evaluation revealed significant performance variations across frontier models (single digits to ~80% accuracy), with notable error modes including tool use failures (36% of conversations) and hallucinations from pretrained domain knowledge, particularly from OpenAI models which hallucinated non-existent insurance products 15-45% of the time.
