Ankur Sinha — Principal UX Designer, Human–AI Interaction
03 — PHILOSOPHY
Stream of consciousness
- I design the seam between humans and AI agents.
- The moment of decision, override, and trust.
- AI claims become trustworthy only when their uncertainty is legible.
- And their basis is checkable.
- Reversibility is the policy axis — not 'safety'.
- Code my own prototypes. The prototype is the design argument.
04 — EXPERIENCE
12+ years, four companies, one through-line.
Engineer-turned-designer. Shipping enterprise SaaS and AI-assisted product surfaces across banking, supply chain, oil & gas, e-commerce, and cloud database tooling.
Jun 2013 – Jul 2014
1 year
IBM India
IT Analyst
Where the engineer-turned-designer story starts. Built systems before I designed them — and that's still how I think.
- Engineering
- Origin
05 — THE LAB
What I build when nobody's asking.
I code my own prototypes — React/TS, SwiftUI, Compose, Tauri, Electron, Next.js. The prototype is the design argument. Hand-off to engineering happens with the contract already in code.
Supporting open-source work
WatchTower
Self-hosted deployment platform
Operator-facing tooling for container auto-updates, multi-node deployment, and guided host operations across your own machines. Integrations dashboard for six interconnected tools (Podman, Nginx, Tailscale, Cloudflare, Coolify, Watchdog) so one operator can see live status and recover any of them.
4★ on GitHub · ships to 6 distribution channels
- Python
- Electron
- VS Code ext
- PyPI
GovLens
Context-aware overlay for government portals
Chrome extension that activates on 25+ national gov TLDs. Surfaces translation (100+ languages), structural navigation, a 0–100 usability score across 8 axes, and a region-aware jargon explainer. Three-engine translation cascade — on-device AI → free Google Translate → premium Claude (BYOK) — shows which engine will answer before you click.
25+ TLDs · 100+ languages · 8-axis usability score
- Chrome ext
- On-device AI
- Claude SDK
EMPATHEIA
Multi-modal AI companion (offline-first)
Next.js + AI SDK PWA. Camera-based facial-expression detection feeds mood-aware tone adaptation. Hybrid intelligence fallback: if the model runtime fails, empathy-map quadrants still update via deterministic sentiment + keyword heuristics — the user never sees a dead surface.
PWA · offline-first · graceful degradation
- Next.js
- AI SDK
- face-api.js
- Ollama
Supporting work
The Human–AI trilogy
Helm, Human-in-the-Loop (Sentinel), and Recourse — three faces of the same problem: AI claims become trustworthy only when their uncertainty is legible and their basis is checkable. Connected code prototypes (React 19 / TypeScript / Tailwind v4) with a deliberate shared vocabulary.
06 — HOW I WORK
Usability is the litmus test, not the checklist.
Accessibility, calibrated confidence, and recoverability aren't accessories. They're how you tell if enterprise software is actually any good.
01
Code my own prototypes
React/TS, SwiftUI, Compose, Tauri, Electron, Next.js. The prototype that makes the design argument unambiguously — handed off with the contract already in code.
02
Reversibility as the policy axis
'Is this safe?' is the wrong binary. What is the recovery cost? — that's the scale, and it's what determines when to pause for the human.
03
Calibrated language over raw percentages
Likely / Unsure / Low — vocabulary the user can actually act on. Numbers progressively-disclose on hover, when they're needed.
04
Outcome-driven over task-driven
Surface the outcome the user actually wants, then let them steer the system there. Workflows are means; outcomes are the end.
Credentials & foundation
HFI Certified Usability Analyst
CUA
B.Tech, Computer Science
BTLIT Bangalore · VTU · 2008–2013
12+ years shipping product
Enterprise SaaS · AI surfaces · Native + Web
07 — STACK & BELIEFS
What I work with — and what I work toward.
The tools are interchangeable. The beliefs aren't. The stack column is the surface; the beliefs column is the substrate.
Stack
- Frontend
- React 19
- TypeScript
- Next.js
- Tailwind v4
- Framer Motion
- Three.js
- Native
- SwiftUI
- Jetpack Compose
- Tauri
- Electron
- Languages
- TypeScript
- Python
- Rust
- Go
- Swift
- Kotlin
- AI & runtime
- Claude
- Anthropic SDK
- MCP
- llama.cpp
- Ollama
- On-device AI
- Design
- Figma
- FigJam
- Storybook
- Token Studio
Beliefs
- 01
Human-in-the-loop
Every consequential action passes through a human before it lands. Not a checkbox — a designed surface.
- 02
Legibility over confidence
The model's uncertainty is part of the interface. If it can't be checked, it can't be trusted.
- 03
Reversibility is the axis
Recovery cost — not 'safety' — is what should decide when an action pauses for review.
- 04
Recourse by default
Every action has a documented, accessible way to undo, appeal, or escalate.
- 05
Interfaces are policy
What a UI makes easy is the team's real position. Surface defaults are governance defaults.
- 06
The prototype is the argument
I write my own React/TS, SwiftUI, Compose. Hand-off happens with the contract already in code.