Steel ai
A 70-year-old steel manufacturing giant (presented under the alias 'Steel.ai' due to a strict NDA) faced a critical bottleneck in its AI transformation. Their core AI tool—referred to here as ‘Calculating’ to protect confidentiality—was designed to estimate steel requirements but failed to gain traction. When engineers tried to audit the AI’s math, the system returned dense, unformatted code blocks, making it impossible to verify data or trust the tool. Simultaneously, the company lacked a unified ecosystem to train staff, track AI product usage, or manage internal tech innovation.
Visual identity
Website design
Work produced during my tenure at Acueducto
Role
Product design
Client
Steel.ai
Date
August 2025
The problem
Despite being ready to scale its AI capabilities, the company faced a critical bottleneck. Their core AI tool—referred to here as ‘Calculating’ to protect confidentiality—was designed to estimate steel requirements but failed to gain traction. When engineers tried to audit the AI’s math, the system returned dense, unformatted code blocks, making it impossible to verify data or trust the tool. Simultaneously, the company lacked a unified ecosystem to train staff, track AI product usage, or manage internal tech innovation.
The challenge
The challenge was twofold, spanning different phases of the project. First, I had a strict one-week sprint to completely redesign the 'Calculating' tool to bridge the trust gap between engineers and AI.
The solution
To meet the intense one-week deadline for ‘Calculating’, I spent a full day interviewing a specialized engineer to deeply understand their evaluation criteria. I designed a new interface that visualizes data like their traditional software, allowing them to confidently cross-reference steel quantities and instantly fix table errors via an AI text prompt.
The ‘Calculating’ redesign earned immediate approval from the engineering team, transforming a frustrating bottleneck into a trusted, high-speed workflow for validating AI estimates.
To ensure consistent deployment across these evolving web platforms, I built a highly scalable Design System based on shadcn.
