Designed an AI-agent powered conversational analytics platform that helps engineers explore complex manufacturing data through natural language, replacing a set of separate dashboards and query tools with one conversational workflow.
“Nexus AI” is a placeholder name for an NDA-protected product. Interface and data shown are illustrative.
Conversational Analytics
Enable engineers to ask questions in natural language and receive contextual, data-backed insights without navigating multiple tools or dashboards.
Agentic Insight Exploration
Guides users through multi-step analysis: specialized agents collaborate behind the scenes, surfacing follow-up questions and pointing to the next area worth investigating.
Conversational Analytics
Users type a question in plain language and get an answer, not a dashboard to configure.
Conversational Analytics lets users explore complex data through natural language instead of filters, charts, or queries. By turning analysis into a dialogue, it delivers contextual, actionable insights faster and with less cognitive effort, matching how people actually think and make decisions.
As AI capabilities advance, conversation is becoming the default interface for analytics and enterprise workflows.
- How might conversational AI support complex analytical reasoning without oversimplifying expert workflows?
- How can agentic systems guide users while preserving user autonomy, trust, and transparency?
- How can conversational UX integrate into existing enterprise tools without disrupting established workflows?
Modern manufacturing analytics involve navigating highly fragmented data across multiple systems, making insight discovery slow and cognitively demanding.
As the Product Designer on this initiative, I designed an agentic, conversational UX layer on top of existing analytics workflows, so engineers can explore data through dialogue rather than manual configuration.
The design centers on human-AI collaboration: the AI synthesizes data and suggests next steps, while engineers stay in control of interpretation and decisions.
This work builds on earlier AI summarization efforts and marks the shift from static insights to interactive, exploratory analytics, setting the direction for future AI-powered manufacturing experiences.
One workflow replaces dashboard sprawl
Centralizes fragmented data, tools, and intent into a single conversational workflow: engineers move from question to insight to action without switching between tools.
Agents that show their work
While agents run a multi-step study, the UI narrates each step in plain language and keeps it interruptible. Users can redirect the plan mid-run instead of waiting for a black-box result.
Progressive depth, not simplified answers
Every answer opens with the conclusion, then lets engineers unfold the evidence: clusters, distributions, and the raw records behind them. Output formats were defined through Golden Set validation with domain experts, so each answer arrives as the text, chart, or table that reads fastest.
Prototyping on the production stack, before any PRD, cut the time between an idea and a stakeholder decision.
Prototype-first, not PRD-first
A working prototype of the 3 core flows in 48 hours, at roughly 80% of the target fidelity, before a single line of the PRD was written. Stakeholders reacted to something real instead of a spec, so the direction locked far earlier.
Faster design-to-development handoff
The Figma mockup and a working prototype, built on the production stack (AngularJS), were developed in parallel and handed off together. Engineers had both the file and working code to reference, not just a redline, cutting handoff time by about 30%.
Shorter concept-validation cycle
The prototype-first lifecycle replaced the previous PRD-first process, compressing a typical concept validation cycle from about twelve weeks to two.
Try the interactive prototype, or see the full process, iterations, and detail in the Figma case study.