Thirteen years designing enterprise software for network engineers at HPE, Juniper Networks, Riverbed, and Cisco. Now applying that same rigor to RAG systems, AI agents, and MCP integrations — one person doing the research, the interaction design, and the implementation.
I'm Melinda Arora, a UX designer turned AI product builder, based in Bengaluru. For over a decade I've led interaction design across some of the most technically dense domains in software — network management, cybersecurity, and enterprise infrastructure — shipping products that engineers actually trust in high-stakes environments.
In 2026 I started closing the loop between design and implementation: learning to build the RAG pipelines, AI agents, and MCP-connected tools that used to sit on the other side of a handoff. I now take AI-powered products from research through to a working build — one person, one throughline, no translation loss between the design and the code.
I'm HFI-certified in usability analysis and experience design, IDEO-certified in Design Thinking, and hold four U.S. patents for interface and security-tooling inventions built during my time at HPE and Juniper.
Internal teams asked plain-language questions against years of scattered docs and got answers with citations back to source, instead of another folder to search manually.
A supervisor agent plans a research task, delegates it to worker agents, and routes the result through a human approval step before anything ships.
Photos and blueprint scans become structured takeoffs and cost estimates for trade contractors, replacing manual line-item entry from a PDF.
A desktop assistant reaches into a team's live docs, tickets, and calendars through custom MCP servers, so answers stay current without copy-paste.
Field research, user interviews, and a scoped problem statement before any architecture gets picked.
Data sources, retrieval strategy, and agent design mapped out — the decisions that are expensive to change later.
Working interface and working pipeline, built together so neither one gets designed in a vacuum.
Usability testing alongside model evals — accuracy and groundedness checked the same way heuristics are.
Deployed, instrumented, and watched — with a plan for what happens when the model or the data shifts.