I’m Aryan Soni. I shape AI products, build useful software, and help teams put it to work. Open to product roles and consulting projects.
I work with users and stakeholders to define what matters, what to measure, and what to leave out. In one program, that meant reducing 32 executive metrics to 10.
I build prototypes and products to test the important assumptions early. The seven projects below cover consumer apps, internal tools, voice AI, and personal software.
I connect product decisions with evaluation, rollout, and adoption. A working product needs a team that can use it, and a way to tell whether it is helping.
I like the work between an interesting idea and a product people can use. Understanding the problem. Testing what is possible. Making the tradeoffs. Then staying close enough to learn what needs to change.
Consumer products, business tools, and experiments. Each case study explains the problem, my role, and the choices that shaped the result.
A bilingual voice agent that calls credit card customers who stalled before activation. She cannot leak a credit limit, because she was never given one. Then I sent seven hostile personas at her and the funniest thing happened. The polite ones broke her. The aggressive ones never did.
Two people, one couch, thirty minutes of scrolling and no decision. So I built the fix: you both swipe, and the first title you both want becomes a match. 90+ real users, real Stripe payments, and one stance I am proud of. It will never charge monthly. Pay for the moment it saves, not the habit it creates.
A drone R and D team was tracking 3000+ parts on spreadsheets, and nobody could say what was in stock or who had it. Engineers lost six hours a week just looking. I ran the discovery, wrote the spec, and shipped in four weeks. A part is never in stock or out. It is in one of six states.
A beverage manufacturer was running its sales pipeline on memory and forwarded WhatsApp messages. I own that whole funnel now: the websites, the SEO that fills them, and the CRM on top. Total infrastructure bill: zero rupees. Then the data said something nobody expected. Energy drinks were half of all demand.
Two games about your senses. Rebuild a colour from memory. Hold a button for exactly 2.7 seconds. Then get roasted by one of 275 hand-written lines that mathematically cannot repeat inside a run. No accounts, no typed names. Nobody can type anything offensive if nobody can type at all.
Act one: a complete subscription tracker. It worked exactly as designed and almost nobody came. Act two: I believed the evidence. People wanted the reminder, not another app to open. So I built the second answer in one day: a tracker with no app at all, living inside the calendar and inbox. One problem, two answers. One is invisible.
A Mac app that answers exactly one question every morning: how much can I actually spend today? No server, no account, no telemetry. My money data has never met the internet. Every finance app wanted my bank login. Five seconds of typing was the better trade. Never store a number you can calculate.
Turn a business problem into a product direction, a testable prototype, and a practical delivery plan. I work across discovery, prioritization, evaluation, and rollout.
Replace manual handoffs and scattered information with tools that fit the team’s process. My projects include CRMs, inventory systems, and automated reporting workflows.
Define what an agent should do, where it should stop, and how to assess its behavior. I bring conversation design, adversarial testing, and structured evaluation to that work.
Connect delivery with how people will use the product. Stakeholder alignment, KPIs, rollout planning, and change management, with clear ownership beyond the pilot.
A smaller scope can protect the part of the product that matters. My case studies include the features I dropped, from bank connections to extra agent tools, and the tradeoffs behind those decisions.
A product can work as designed and still miss what people need. Subdue taught me to revisit the format of the solution, rather than keep adding features to the original idea.
I look for ways to reduce risk through the product itself: keeping financial data local, limiting agent context, and avoiding unnecessary data collection. These choices support testing and safeguards; they do not replace them.
Sometimes a spreadsheet is the right foundation. Sometimes the product needs a native app or an AI agent. I choose based on the team, the constraints, and what it will take to maintain the result.
I'm the friend who has already tried the app you just discovered, rearranged his homescreen so information travels faster, and set up three AI workflows before breakfast. I fell for technology in the early 2000s, somewhere between a big box computer and a stack of CDs, and I never recovered.
Work-wise, they call it product management. Really it's this: I listen to what people actually need, then build technology that looks amazing, feels amazing, and gets out of the way. I fix messy processes for businesses, untangle the problems people have with their tech, and ship my own products when an idea refuses to leave me alone. My whole business philosophy fits in one line: create real value first, and the rest follows.
America shaped me for three years: a Master's at UT Dallas, President of the Entrepreneurship Club at the Blackstone LaunchPad, and every founder event around Dallas, where I learned my most valuable professional skill: showing up. Sometimes even invited.
These days I work with global companies by day and build my own products by night. I've shipped AI for billion-dollar enterprises, and an app so two people can pick a movie without fighting. The movie one was harder. Somewhere in between, Emergent ranked me among India's Top 300 Vibecoders, out of 20,000+ applicants.
If you've read this far, we would probably get along.
If you’re hiring, tell me about the team and what you need this person to own. If you’re exploring a project, tell me what is getting in the way. We can start there.