Skip to content
Abhishek Anand

Abhishek Anand

UX Engineer at Google and author of Designing Human-Centered AI Interfaces. Building agentic AI tools, full-stack systems, and the data pipelines underneath them.

I build tools that help researchers find answers and teams understand their data. Here I share the work, the decisions behind it, and what I learn along the way.

Abhishek Anand

Currently in Bengaluru

Working at Google · Open to conversations

Cover of Designing Human-Centered AI Interfaces by Abhishek Anand

New book · Available now

Designing Human-Centered AI Interfaces

Building understandable, trustworthy, and controllable AI products

Sixteen chapters of interaction patterns for probabilistic products: inputs, outputs, explainability, steerability, memory, agents, and the handoff to engineering. Paperback and ebook from BPB Publications.

See the contents

Selected work

Projects I’m proud of

See full archive

Google Ads Central

In production at Google

Agentic research tooling

Agentic desk-research tooling for UX researchers. The system pulls together third-party reports and analyst data, then reasons over them with a multi-step agent. Built on the Agent Development Kit with Gemini, A2A for delegated subtasks, and A2UI streaming the agent's steps as it works. I built the ingestion pipelines, the agent orchestration, and the front-end.

ADK · Gemini API · A2A · A2UI · Python · Data Pipelines

Read the architecture notes
Conceptual workflow
  1. Reports & analyst data

  2. Multi-step research

  3. Synthesised findings

Google Ads Central

Live at Google

Internal ads analytics hub

Led development of the analytics hub Google Ads Central UX teams use to explore research and product data in one place. Angular on the front, GoogleSQL underneath.

Angular · JavaScript · GoogleSQL · Material Design

Read my Angular architecture notes

Google Ads Central

In production at Google

Natural-language data exploration agent

An agent that lets anyone query a large internal ads dataset in plain English. It finds the right tables through retrieval over curated schema knowledge, writes and runs the SQL through tool calls, then explains the result with charts and full data lineage. I built the retrieval, the orchestration, the evals, and the UI.

Gemini · RAG · NL-to-SQL · GoogleSQL · Data Lineage

Conceptual workflow
  1. Plain-English question

  2. Schema retrieval & SQL

  3. Charts & data lineage

Google Assistant

+20% daily active users

Duplex for Web: Brazil General Elections 2022

Built an assisted workflow that helped Brazilian voters find their polling booths during the 2022 general elections. Daily active users grew 20% during the launch window.

JavaScript · Web APIs · Accessibility · I18n

What I do

A short list of the things I’m good at, refined over 17 years

Agentic AI engineering

Multi-step agents on the Agent Development Kit, Gemini for reasoning and synthesis, A2A for splitting work across specialised agents, RAG with hybrid retrieval, NL-to-SQL, the ingestion pipelines that feed them, and the evals that keep them honest. Most of my recent work lives here.

UX engineering

The seam between design and code. I write the interfaces, the interactions, and the thousand small defaults that decide whether a product feels considered or not.

Full-stack engineering

Comfortable shipping from the database up through the layout: JavaScript and TypeScript on the front, Node and Python on the back, with the data plumbing in between. I like the parts where systems meet each other.

Architecture & technical leadership

Designing systems that are honest about their tradeoffs, and helping the engineers around me make good calls when no one is watching. 17 years of doing this across startups, scale-ups, and Google.

A working note

I'm most interested in the parts where systems meet each other. The seam between design and code, between agents and the people they serve, between the messy data and the clear interface that finally makes it useful.

Working principle, since 2008

Get in touch

Say Hello

If you want to talk about speaking, the book, or compare notes on agents and interfaces, my inbox is open.