I take complex processes and make them simple.

The problem

At Mobideo, a B2B platform for managing complex industrial operations, a core workflow was spread across eight different users - each handling their own tasks, checklists, and file uploads, with no single place to see or coordinate the work. It was slow, people-heavy, and error-prone.

My role

I led this end to end - driving the design and build largely on my own using AI tools, with some support from a developer. I owned the workflow design, the dashboard UX, and the integration of AI agents into the process.

Key decisions

• -Consolidated a multi-step, multi-user process into one dashboard rather than improving each step in isolation.
• -Embedded AI agents directly into the workflow so the system handled more of the work automatically.
• -Designed for fewer, more empowered users - a deliberate bet on efficiency and cost savings.

AI quality & iteration

To make sure cutting the team in half didn't lower quality, I built an evaluation and feedback loop into the process. Agent output was checked against expected results, and reviewer corrections fed back as a source of improvement - so the agents got more reliable over time. The dashboard made it clear what the AI had done versus what needed a human, which kept users trusting the system instead of re-checking everything.

The problem

Mobideo customers need to turn manual instructions and procedures into digital checklists to use them in the platform. Doing it by hand could take weeks, and in many cases the output has to meet FDA standards - so accuracy is non-negotiable and every error carries compliance risk.

My role

I designed the AI workflow end to end - defining how the system interprets each customer's needs, manages context across the conversion, and produces output in the exact format they require.

Key decisions

• -Built it as an agent-based process with deliberate context management, rather than a single prompt, to handle long and complex procedures reliably.
• -Designed around a human-in-the-loop model: AI does the heavy conversion, people review and validate.
• -Made output adapt to each customer's desired format instead of forcing one rigid template.

AI quality & iteration

Given the FDA-grade accuracy requirements, getting the conversion right mattered more than getting it fast. I tested the workflow on a set of ~50 sample procedures and ran an iterative loop with subject-matter experts: SMEs reviewed the AI's output, flagged what was wrong or imprecise, and I used their feedback to refine the prompts and the context the agents worked from. Each round tightened the output - while the SME validation step stayed in place as the auditable human control that kept the process compliant.