Sudhanshu Gautam · Lead UX Designer, Wikimedia Foundation
I'm a human-centred designer: I care about the people problem first. I adapt quickly, and right now I'm really enjoying exploring what AI makes possible.
Lead UX Designer at the Wikimedia Foundation, on the Language and Product Localization team. Before that, nearly four years at Noora Health, where I co-led design on remote health engagement tools that help families continue care after leaving the hospital, and CIID, where I first learned to think like a designer. I started out as an engineer, with about four years at Tech Mahindra.
How I work
I've designed for first-time internet users on feature phones, families continuing care after a hospital stay, and volunteer translators across hundreds of Wikipedias. The research comes first; the screens come later.
I'm comfortable in the terminal with Claude Code, Codex, Grok Build, Gemini CLI, Cursor, OpenCode, Copilot CLI and Pi, and I wire agents together when a job needs it. When a new model ships, I try it on real work.
I'm really enjoying exploring AI. I build small things to learn, and some become real products: Promptbook went from first commit to the App Store in 25 days. I share what worked, and what didn't, on YouTube.
Wikimedia Foundation
Lead UX Designer since 2020, now on the Language and Product Localization team.
downloads on KaiStore
A Wikipedia reading app for smart feature phones. Designed keypad-first navigation and reading layouts for 2.8" non-touch screens, targeting first-time internet users in emerging markets.
stories created, 89 returning editors
A visual storytelling format for Wikipedia. Turns articles into short, card-based stories using images from Wikimedia Commons. Designed and piloted with the Indonesian Wikipedia community through workshops in 4 cities.
Wikipedia language editions
A tool that helps Wikipedia editors translate articles across languages with integrated machine translation. Supporting UX for the translation workflow, working with volunteer translators across language communities.
Projects
Apps, tools and a channel I've made on my own. Open any of them for the full story.
iPhone + Mac apps · 2026
Your iPhone, as your Mac's camera. Plug it in with one cable and look great in Zoom, Meet, FaceTime and OBS.
Mac webcams make people look worse than they really do. Lenscord turns the iPhone you already own into your Mac's camera, plus the tools to record like a pro.
It's two apps. The iPhone app captures and encodes the video. The Mac menu bar app hands it to video calls as a camera called “Lenscord”, and to OBS as a 4K source. The live video only travels over the cable to your own Mac. Solo: product, design and build, with Swift and Claude Code. iPhone build 27 went to TestFlight on 25 October 2026.
The panel shows one status word, at most one button and three switches. Everything else lives in Settings. Setup goes Connect, Pair, Choose and Finish, and each step moves on by itself as soon as it works.
One status word, three switches, then the tools.
Type the iPhone's six-digit code once. After that, the two connect by themselves.
Video calls and other apps, OBS, or both.
It scrolls as you talk. Speech is recognised on the Mac, never on a server.
Background blur and Darken background run on the iPhone.
Mac screens from lenscord.com, with demo data. The product shot above is built from Lenscord's own marketing art: the person is an illustration.
One camera screen with two modes, Mac camera and Shorts, above a shutter row that never moves, like Apple's Camera. Opening it never starts the camera. In Shorts, the script sits right under the lens and waits when you pause.
iPhone screens from the current build and from lenscord.com, with demo data.
The iPhone, the Mac, the website and the docs share one small set of words. Status is always one short phrase from the same list, and some words never appear on screen at all.
If the cable wiggles or a call comes in, OBS keeps the last picture and carries on in the same recording. Backups are written in parts, so even a crash leaves a playable file.
When the phone gets hot, Lenscord turns effects off first and lowers the frame rate only as a last resort, never the picture size, and both apps say so in one line.
Apple's Continuity Camera is free and built in, so the site compares all three options with sources: “For quick calls, it's great. … Lenscord is for when you're making something.”
iPhone app · 2026
Say it once. Use it forever. A private home on your iPhone for the AI prompts you reuse.
Many of the prompts we give AI tools get used again and again: who you are, what the project is, how you want the answer written.
Promptbook keeps them in one private place. You talk, it writes your words down on the iPhone, and one tap copies them, ready to paste into any app. Solo: product, design and build, in Swift with Claude Code and no third-party dependencies.
From the App Store listing for version 1.0.1.
Speech recognition runs on the device only. A check in the build fails if the app's code ever references a network client.
An early build had a custom keyboard. It couldn't actually type, which put App Review at risk, so I removed it. Copying works everywhere a keyboard would.
Where the recogniser takes about a minute at a time, recordings are split into 45-second chunks that overlap by two seconds, and the repeated words are stitched out.
Saves are atomic and keep the previous copy. A book that can't be read gets a recovery screen instead of being overwritten.
Recording uses the built-in microphone and skips Bluetooth headsets, whose call-quality audio is worse for recognition.
New prompts sit on top and the ones you use most rise, so the right card is usually already in reach.
Version 1.2 went to TestFlight on 26 October 2026. Start from the app icon, Siri, a widget or Control Center, and it opens already listening. When you stop, your words are already copied, and one tap opens ChatGPT, Claude or Gemini.
Screens from the 1.2 build in TestFlight.
Video editor · 2026
Raw footage in. A finished short out. The editor I built to turn my recordings into YouTube videos and Shorts.
I record in Hinglish and English: a 4K talking head in OBS, clips from my iPhone, and screen recordings of the tools I'm testing.
Vedit turns that into a finished 16:9 YouTube edit plus vertical Shorts and Reels: cut, captioned, colour-graded, scored and checked. As a Claude Code skill, Claude makes the editorial calls. Vedit Studio, the Mac app, has an Autopilot that runs on rules, my choices and local models like Whisper, with no language model at all.
Each style is a full visual system with its own captions, graphics and sound: word-pop captions and stickers, a terminal look, explainer cards that build up, and a race between three AIs with live timers.
Clips from reels Vedit rendered, muted here.
It cuts retakes and fillers, writes Hinglish captions that pop with my voice, and hides emails and OTPs with masks that follow the screen. Then it grades, adds music and sound, and runs a QA check on lip sync, loudness, captions and privacy.
A QA board from a sample edit: faces, safe areas and text sizes marked on every beat.
Projects point to the footage, and the originals are never modified or uploaded.
Credentials, codes, personal details and other people's chats are masked for the whole video.
Every claim on screen has to be true, and visible or said in the footage.
YouTube channel · since May 2026
A designer trying AI agents, design and coding tools on camera, in English and Hinglish: what worked, what didn't, and what I'm still figuring out.
“I'm not an AI expert. I'm learning by doing: building small things, testing new workflows, and finding useful ways to bring AI into work and life.” That's the channel's own description. Counts are from October 2026.
Gemini 4 Argon Review: Ek Photo Se 3D Game Ban Gaya (Hindi)
AI meri job le lega? Ek product designer ke honest two cents
Maine apni YouTube videos AI se edit karwayi | Claude Opus 5.5
Grok Build ne meri iOS App bana di! Build Fail se Working App tak
Claude Fable 5 Just Redesigned My Real Website (Honest Test)
Grok Build CLI: A Designer's First Look
Codex CLI ke 5 naye updates: ab ye BAAT bhi karta hai
Superfast Review of DeepSeek Harness on MacBook
Website · 2026
Does your software have a CLI? A verified directory of 78 developer tools, with install commands and the publisher's own docs.
When you work with coding agents, it helps to know whether a product has an official command-line tool, what the command is called and how to install it safely.
whichcli answers that for 78 developer products, from Claude Code and Stripe to Figma and Docker. Every answer links to where it came from, and the site looks and behaves like a terminal you can browse from the keyboard. A solo project; AI agents keep the records up to date, and I'm accountable for them.
Records show status, executable, install command and the date verified, with primary sources.
Figma's CLI is marked “Official · scoped”, because it “is not a general Figma file or design CLI”.
All 78 entries stay readable with JavaScript off. JavaScript only adds the interactive previews.
Website · 2025 – now
India's specialty coffee, made simple. Three taps, no quiz: how you brew, a roast and a budget narrow about 960 coffees to a few.
I built Indian Coffee Guide to make it easy to discover Indian specialty roasters and their coffees in one place.
Product details come from the roasters' own websites. Prices are shown per bag and per 100 g, checked every day, with a list of what got cheaper. Every roaster has a page, and each coffee links straight to the roaster's shop. Built with Next.js, Supabase and Vercel; numbers from October 2026.
City, founding year, active coffees, roast spread and price range, with a link to shop direct.
The same three-tap finder.
Skills, plugins and widgets
Keeps the agent's chat on the left and opens mockups in a pane on the right, inside Kitty. Works with Grok Build, Claude Code, Codex and Cursor.
Start it with /timer 1h. It reads my Google Calendar to know what I'm working on, plays binaural focus audio, and checks in every 15 minutes.
The skill behind Vedit. Claude edits my raw recordings into finished YouTube videos and Shorts, and only stops to ask about real blockers.
Makes episodes of Pip, a toddler cartoon about a turquoise baby dinosaur and his snail friend: story, keyframes, voice, animation, music and QA. The visuals are AI-generated. A first 34-second sample is made.
Read-only access to Wikimedia Phabricator for coding agents: look up tasks, search, browse related tasks and view attachments.
Search Commons for freely licensed images and drop them into your designs, or fetch Wikipedia article content with its formatting intact: bold, italic, headings, links, superscripts and subscripts.
Test UI layouts in other languages before shipping. Select a frame, pick a language, and the plugin translates the text in place with AI. Smart filters skip buttons, URLs, dates, placeholders and component instances.
Browse and resume past AI coding sessions from Claude Code, Codex, OpenCode, Copilot, Pi, Gemini and Grok, right from the bar.
A calm Wikipedia reader for while an agent works. Pick a mood and follow a good rabbit hole without losing sight of the work.
Every SQLite database touched in the last 60 days, at a glance. Click one to open it in DB Browser.
What runs next on this machine, and whether the last run worked. systemd timers and cron, read-only.
I'm happy to talk about designing for people, AI agents, or anything on this page.
Sudhanshu9252@gmail.com
