I'm Nikita Kharya — a UI architect and principal engineer. For more than twelve years I've turned complex enterprise systems, across banking, cloud, and fintech, into interfaces people can actually use. That same craft now shapes how people and machines work together.
Coming from a UI background, I've always believed that design is how people make sense of complexity.
That's what drew me to AI — because right now, AI is complex in ways users can't see. And I believe thoughtful interface design can actually shape the way humans and AI collaborate.
Interface architecture
Folded three separate applications into a single workflow — one surface for an operation that used to span many. Built on a shared design system and micro-frontends.
A framework where JSON schemas generate entire interfaces — powering transaction-monitoring and compliance tools, and cutting manual review risk across multiple banking apps.
A reusable component library serving teams across the org, plus a React migration of the CloudHealth UI for multi-cloud analysis — build the system once, and everyone ships faster.
Designed how the team builds with AI: precise specifications feeding AI-assisted development, compressing three-week cycles into days — without giving up rigour or control.
At the human–AI seam
An interface that turns a model's static chain-of-thought into something a person can navigate — raising verification accuracy and lowering cognitive load. Interface design as AI research.
A UI for agent transparency — Timeline, Graph, and Compact views that let you watch an agent think, with an approval gate to confirm what it intends before it acts. Adapters for Vercel AI SDK and AG-UI.
A quick illustration of the human-in-the-loop pattern — reasoning made visible through progressive disclosure, with a human checkpoint placed exactly where the stakes are high. The real components live in the agenttrace-ui demo (library on GitHub) — the same ideas behind my ACM IUI 2026 paper.
Turning three-week features into three-hour pull requests with spec-driven prompts — and the verification discipline that keeps it safe in regulated environments.
Read on Medium ↗As agents act on our behalf, the interface around them hasn't kept up. The reasoning gap, the handoff problem, trust calibration — and what agent-aware chrome should look like.
Read on Medium ↗Why agentic systems need a dedicated interaction layer — the interface between people and the agents now acting on their behalf.
Read on Medium ↗Agyeya makes AI interpretability and safety research open to anyone — no institution required. We run free workshops with no barrier to entry, connecting students who have the talent but not the access to real, open research.
agyeya.com ↗
Open to interface architecture, human-AI collaboration, research, speaking, and advisory.