I take AI products from idea to live.
Recently I shipped two end to end — NoSubs and Quesify — both live and in users' hands. Below: what I built, how I think, and a live teardown you can try on any product.
Two AI products, both live.
Quesify LiveThe AI chatbot that speaks first — for small-business sites.
Most visitors hesitate on pricing or booking, then leave without a word — on Quesify's own numbers, about 96% of hesitating visitors leave silently, no question asked. Enterprise chat (Intercom, Drift) is too heavy and pricey for a café or clinic where the owner is the entire support team.
Detect hesitation — pricing revisits, CTA dwell, exit intent — and open the chat first. Auto-build the knowledge base from the site's own pages so setup is ~60 seconds. Push a notification to the owner the moment a real customer needs a human.
Watch for pricing revisits, CTA dwell, and exit intent — the tells of a visitor about to leave without asking anything.
Open the chat with a line that matches what the visitor is actually stuck on, not a generic “Hi, how can I help?”
Owner pastes their URL; Quesify crawls the site's own pages and is answering questions from it in about 60 seconds — no FAQ spreadsheet, no prompt-writing.
When a visitor is frustrated or asks something the bot can't close, push a notification to the owner with the visitor's exact words so a human can jump in mid-conversation.
Product lead — product strategy, feature scope, AI + push-notification handoff.
Quesify is live and taking its first businesses now. What I'm watching once real traffic lands: what share of hesitating visitors get caught before they leave, and how often a push-alert rescue turns into a booked customer.
NoSubs LiveStop paying for subscriptions you forgot.
People quietly leak money on subscriptions they signed up for once and never cancelled. Manual trackers die because nobody keeps them updated, and the apps that do it for you often take a cut of whatever they save you.
Detect recurring charges automatically by scanning email, then pre-empt the next bill with a timely alert. Zero manual entry — the product earns its keep on day one.
Link Gmail or Outlook, or upload a bank statement — NoSubs never sees a bank password.
Scan for recurring charges other trackers miss — free trials, PayPal charges, annual plans — not just the ones a linked bank account would show.
Get alerted before a renewal hits, then keep, track, or cancel with one tap; NoSubs drafts the cancel email so it sends straight from the user's own inbox.
Product lead — concept, scope, detection design, launch.
NoSubs is rolling out market by market, starting with Australia. What I'm watching first: how many forgotten subscriptions the average connected account surfaces, and how many of those turn into an actual cancellation rather than an alert someone ignores.
How I judge an AI feature.
A sample of the first-pass teardown I do before building anything AI-powered — who the AI is really for, whether it earns its place over a simpler feature, where it can fail, and what I’d ship next to close the trust gap.
AI answer engine that cites its sources.
Perplexity's real product isn't the model — it's the citation. Every claim links to a source the user can check, so trust doesn't rest on the model's confidence alone. That's the bar I hold any AI feature to: if a user can't verify it, it doesn't earn the word 'answer' — it's a guess with good formatting.
How I work.
I start at the moment users give up — the forgotten charge, the silent exit. That's where the product lives.
AI only when it removes real work — auto-detection, auto-knowledge-base. Never a chatbot for its own sake.
Real product, real users, fast. Both NoSubs and Quesify shipped lean and learned in the open.
Zero setup, one-line embed. If a user needs a manual, the product isn't finished yet.
The latest in AI, made clear.
The most important AI news, in plain language — refreshed regularly. Pick a genre and skim the latest, about 50 words a story.
OpenAI open to slowing down AI development
Sam Altman told employees OpenAI would consider slowing its pace as concerns grow over increasingly capable models — after a run of incidents where autonomous agents acted in ways researchers didn't expect. It's a rare public admission that capability may be outrunning safety controls.
AI shouldn't only serve the well-funded.
The same systems that grow startups can help a charity reach more people with less. Below is a sample of the concrete, low-cost AI ideas I put together for a non-profit — ranked by effort and impact.
Turn each session's notes into a short, friendly progress summary in the parent's own language, sent automatically. Keeps families engaged without adding admin.
Suggest books matched to each child's level and interests so volunteers spend time reading, not searching shelves.
Draft first versions of grant applications from your program's own data, so a small team can apply to far more funders.
Nikunj Kaushik
I'm a product manager who ships. While most AI talk lives in slide decks, I build — two live AI products, both taken from a blank page to real users by me alone.
I'm drawn to the quiet moment where people give up: a subscription they forgot to cancel, a website visitor who hesitates and leaves. That's where I point AI — not a chatbot for its own sake, but detection and automation that removes real work. NoSubs reads your inbox and catches recurring charges before they hit. Quesify notices a hesitating visitor, answers from a site's own pages, and pushes the owner a notification when a real customer needs them — live in about a minute.
What's rare isn't any one of those skills — it's holding all of them at once: spotting the problem, judging whether AI actually earns its place, and shipping the thing myself. I teach what I learn, and I don't think AI should only serve the well-funded — which is why I build for small businesses and non-profits. I'm looking for AI product work where shipping and judgment matter more than headcount.
Two live AI products conceived, built, and launched by me — not features I influenced inside a big team.
I use AI where it removes real work — auto-detection, auto-knowledge-bases — and cut it where it doesn't.
I teach it, explain it plainly, and apply it to real missions — from small businesses to non-profits.
The short version: while paying the bills in sales, I taught myself AI and shipped three products — Feedzup, NoSubs, and Quesify. The sales years built the customer instinct; the nights and weekends built the products.
Built and launched NoSubs and Quesify while sharpening people skills in car sales. Learned AI by shipping it, not just reading about it.
Built an AI-powered news aggregator from concept toward launch — led product scope and AI summarization design. Project dissolved due to internal team conflict — a hard lesson in co-founder alignment.
Face-to-face selling — reading real buyers, handling objections, closing. The customer instinct most PMs never get to build. Built Feedzup on the side during this time.
Selling over the phone: clear messaging, fast trust, and hearing what customers actually want.
Design work across a range of client projects.
Designed digital products inside a global consultancy.
Led UX work for one of India's largest telecoms.
Designed customer-facing telecom experiences at scale.
Designed and built web pages — where the technical grounding started.
Building something with AI?
I’m open to AI product roles and interesting problems. Tell me what you’re building and where it’s stuck.