Staff / Principal Product Designer · AI & LLM Products · fully remote (GMT+5:30)
Your model is right. Your users still won’t bet on it.
I design AI products that professionals actually trust and act on. Sixteen years, five industries, one problem — and the receipts for all of it.
Talon Outdoor — six AI systems · PE & VC platforms · PTC — NASA, Boeing & Apple among clientele · Telefónica O2
What changed when they trusted the machine.
Receipt — campaign planning, 2 wks → 3 hrs
Receipt — deal screening, 60% faster
Receipt — $1M / yr, and a business model shifted
Receipt — 4M+ users, co-designed then coded
Selected work — seven products
The products behind those numbers — the first three are AI, two of them LLMs reading unstructured documents. Open any door; where a client is under NDA I walk the real artifacts on a call.
An Ad Agency Became the Market’s Aggregator
Traders watched an algorithm beat them and played their hunches anyway. The fix was letting them argue back — plans they could reshape in one click, every edit teaching next week.
Deal-Screening AI That Cites Its Sources
Analysts get paid to doubt confident numbers. So it shipped already able to name a source for every claim, and to say “I’m not sure” instead of bluffing.
Due Diligence You Can’t Rubber-Stamp
Partners stake millions on claims they’ll never personally verify. So I made them do more work: sign-off stays locked until every signal’s evidence has been opened.
The range under it — four more products, same job
Numbers survive scrutiny; a person ships them. Who you’d actually be getting, and how I run a room.
By Friday of week one I’ve read your evals and sat in on your customer calls; by launch, the interface I drew is the code I shipped.
I read the model at the eval layer. I shape what gets measured. I ship the front-end.
On my first accessibility project I spent a week with my monitor switched off, navigating by screen-reader — designing for someone who isn’t you starts by becoming them. That project started as my documented miss →
Before rebuilding this site I went looking for the study behind the famous six-second portfolio scan. It does not exist — the closest real figure is a 2012 eye-tracking study of résumés. What twenty-three practitioner and eight academic sources actually describe is four readers with four different budgets, and this site is built for them. The audit, and what I changed → (opens in a new tab)
Candidate Facts
Serving size: 30 minutes · Servings: 1 seat
- Experience
- 16 yrs
- Industries
- 5
- Patterns in production
- 8
- Act / Review / Ignore
- 1 rule
- Users shipped
- 4M+ · public
02 · 2How I lead — come argue with me
I lead the way I design: evidence before opinion, overrides as signal.
With engineering
Eval design before interface design
The front-end I own ships in the PR — commented and ready for review. When we disagree on feasibility, the cheapest experiment goes first and settles it.
With product / founder
Outcomes, not deliverables
I’ll contest the roadmap when the numbers contradict it. The design doc is mine to write; the spec is yours.
With customers
Five calls in week one
One a week forever after, and I read the raw support tickets myself. No AI feature ships until I’ve personally watched someone fail to use it.
My first accessibility pass buried screen-reader users in verbose labels. I spent a week with the monitor off, navigating by screen reader, then rewrote the front-end. The full account, in the PTC case →
Allocating creative across a campaign’s hours: I proposed a simple table of input fields. The product owner wanted an interactive time-distribution map; I argued it was too expensive for the MVP and lost. An engineer messaged me late at night saying we should push for it and he would build it — so I designed it over several late evenings and the team shipped it. It became a demo selling point, and was sold as a differentiator. I was wrong about the ceiling, and the person who corrected me was on the other side of the handoff.
A language model has no edges. So I draw them.
Ask one anything and it answers — fluently, in or out of its depth. A user’s trust does not erode slowly; it dies in a single confident wrong answer. So before launch I write a capability contract — a one-pager the whole team signs — and it goes in the interface, at the point of use.
In scope
It’s for
The job stated narrowly — “score deal risk from the data room,” not “analyse documents.”
Declines
It can’t
A named limit, not a hedge. “I can’t price illiquid assets” — never “results may vary.”
Hands back
Ask a human
A visible route to a person, designed with the care of the happy path. Never a silent, confident guess.
The full library — the Act/Review/Ignore rule and 8 patterns, citable →
live — the contract, workingsigned
It can’t: price illiquid assets.
Hands back: anything else → your analyst.
answers openly canned — the contract is the demo, not a model
03 · 2How I build with AI — the workflow, not the tool list
The design system became a plugin. The quality bar became code.
The system, installed
Design systems as Claude plugins
A client’s design system encoded as an installable plugin, so every generated screen speaks the system — two shipped across client orgs, one public. Talon’s five-product system runs this way, installed by developers at Talon and Sahaj.
The bar, automated
Quality gates in CI
Contrast, overflow and legibility checks run on every change to the rendered site before a human reviews it — the same gates that ship this site. When a gate and my eye disagree, the pixels win.
The prototype, shipped
Working code, to production
Prototypes are working HTML/CSS/JS from the start — usability-tested as real code, and that same code is what goes to production.
The judgment stays mine: no AI feature ships until I’ve personally watched someone fail to use it. The receipts are below — this site, torn down →
Proof, not philosophy: loop.js — 42/42 tests · trustlint — 7 rules · this site — 0 build steps
A client CTO and the tech lead who sat beside him for five years, an engineer, a manager, a designer — then three essays, and one form.
Don’t take my word for any of this.
The engineer built the platform, the manager ran the team, the designer sat on the next desk — none of them owes the others agreement.
“Over the past four years at Talon, Arpit has been instrumental in shaping four distinct products from the ground up. His user-focused designs are remarkably intuitive yet adept at handling complex workflows… If you need a designer who excels at combining strategic vision with practical execution, Arpit is the person to call.”
“I worked with Arpit for over 5 years as Tech Lead on Talon’s Atlas and Plato platforms… Arpit consistently asked the right questions early, translated ambiguous business goals into concrete user flows, and turned those into rapid, testable prototypes well before requirements were fully locked down. That speed let engineering and product validate direction early instead of building in the dark.”
“Arpit consistently demonstrated exceptional speed, creativity, and attention to detail… his ability to present multiple design options along with clear pros and cons made it much easier for different stakeholders to make informed decisions and align quickly.”
“Arpit has worked with me for years and I value his honesty and hard work. He’s been an integral part of my staff… involved in all facets of the team, from design to development to hiring and onboarding of new members.”
“Arpit teams up with designers very well, not only does he flawlessly execute the UI implementations but he pushes back on design decisions using his UX expertise… I’d recommend Arpit to any team looking to improve their final product.”
These are excerpts. The full set, unedited, on LinkedIn ↗ (opens in a new tab)
04 · 2Writing — three recent pieces
Three recent pieces on making AI products people actually act on.
Shorter pieces go out through my newsletter on Substack (opens in a new tab).
Nobody Reads Your Portfolio. Fifteen Strangers Skim It.
The six-second scan is a myth — I traced it to a 2012 eye-tracking study of résumés, not portfolios. Twenty-three practitioner sources and eight academic ones describe four readers instead, and I rebuilt this site around what they actually say.
(opens in a new tab)The Prompt I Paste Before Every Single Prototype
It says almost nothing about the product. Why the order of your brief matters more than its words — and what I put in before the thing I am actually building.
(opens in a new tab)The Agentic MVP: Why Your Next Launch Will Be Lovable, Not Just Viable
Every founder knows the pit in their stomach on Launch Day. How the rise of agentic systems is rewriting what “minimum viable” means — and why lovability is now the bar.
(opens in a new tab)04 · 3Contact — available · 4 weeks’ notice
One seat. Full-time. Yours to offer.
Staff / Principal product design, or a founding design lead. Fully remote, GMT+5:30, four weeks’ notice.
Indore, India · fully remote · daily overlap with both US coasts · 4 weeks’ notice
Book the 30-minute call ↗ (opens in a new tab) · Resume (PDF) ↓ · Email · LinkedIn ↗ (opens in a new tab)
Rather read first? The pattern library, the free resources, or paste the role into the fit check.
— Arpit
P.S. If you scrolled all the way here: the part I’d most like you to remember is the errata. Everything else is receipts; that part is character.
Gate open — book the 30-minute call ↗