Hi, I'm Parth.AI Product Lead at uEngage · Chandigarh, India

I build AI products for people who are hungry and in a hurry.

I lead product at uEngage, where more than 190 restaurant brands take orders through the screens I look after.

Parth Khungar
  • +145%uEngageKiosk sales, after uKiosk 2.0
  • ~90%PharmaSecureLess time to flag a suspect code, in testing
  • 99%CSIR-CSIOModel accuracy on identical machines

Work

Three problems, and what I did about them.

uEngage · AI Product & Experience Lead · May 2026 – now

When I joined uEngage, almost every request became a yes. Whether it solved the problem, or could ever become a product, came later.

I lead product for customer experience: every way a diner orders, on App, Web, Kiosk and WhatsApp, for 190+ restaurant brands. I started by shipping uKiosk 2.0 myself in about three weeks. Then I changed how we decide what to build, turned my UX team into product owners, and brought AI into the product.

My part

  • Led uKiosk 2.0 solo, my first project: the product, the UX, the UI, some of the code, and analytics built from zero. Kiosk sales are up 145 percent.
  • Became the single front door for requirements, and decide what gets built by whether it solves the problem and can become a product.
  • Turned the UX team into product owners who ship code, custom websites and client solutions, on Claude Design instead of Figma.
  • Brought AI into the product: AI picks, our smart upselling algorithm, lifts average order value by 34 percent where it runs.
  • Trained sales and account managers to present every feature as the answer to a client's problem.
How I thought about it
  1. Ship something firstIn a new role, trust comes from a result. I led uKiosk 2.0 myself, end to end, and it was live in about three weeks.
  2. Measure before claiming39 funnel events and weekly dashboards, so every kiosk claim has a number behind it.
  3. One front doorEvery request now comes to one place, and answers two questions before it is built: does it solve the problem, and can it become a product?
  4. Grow the team with the companyDesigners who drew screens now decide journeys and why, solve for clients and ship code. The team has to grow, not just get bigger.
  5. Close the loop with clientsA feature a client does not understand does not exist for them, so sales and account managers learn each one as the answer to a problem.
  6. Put AI where it paysAI picks ranks by what completes a cart, not what is popular, and lifts average order value by 34 percent where it runs. Inside the team, AI builds the tools that make everyone's work easier.
Illustration
  • 39funnel events, built from zero
  • +145%kiosk sales, after uKiosk 2.0
  • About 3 weeksfrom agreed to live

Every couple of seconds, somebody taps place order.

0orders placed since you opened this page

  • App
  • Web
  • Kiosk
  • WhatsApp
  • 190+ restaurant brands
  • 4 ordering channels
  • 1st product hire

An estimate, from our monthly orders across the four channels, counted over the twelve hours a day restaurants are busiest: about one every two seconds.

+29%

Monthly GMV, with orders up 18 percent, since I joined.

Read the full story

PharmaSecure · Associate Product Manager · Sep 2024 – Apr 2026

Counterfeit medicines were turning up across India. One code had been scanned about 3,000 times.

A top manufacturer could see counterfeits turning up but not the pattern. I picked that problem up in a client meeting and built PS AnalytIQ from scratch, end to end: a hybrid algorithm that learns each manufacturer's normal and flags what breaks it. My mission was to stop counterfeits from reaching consumers, and in pharma a wrong call costs either way: a missed fake can reach a patient, and a false accusation puts a manufacturer's name at risk. So every alert shows its data, in real time, on the manufacturer's dashboard. The manufacturer cracked down, and has since busted several illegal manufacturing units itself.

My part

  • I took the problem on from a client meeting, and built it myself as both the product manager and the engineer.
  • I changed the question: from judging codes to judging scanners.
  • I designed it as a white-label hybrid algorithm that learns each manufacturer's normal.
  • My mission was to stop counterfeits from reaching consumers, so I built every alert to show why it was raised.
How I thought about it
  1. A product for many manufacturersSeveral very large manufacturers were on the platform, and every supply chain is different, down to each medicine. So the approach is white-label, but no single model fits all: for each manufacturer I tested which one fit best.
  2. A hybrid: learn normal, flag what breaks itAn algorithm learns the scan pattern of each manufacturer and medicine, including how long a code usually takes to be scanned for the first time, and flags what departs from it.
  3. Rules for what cannot happenSome things are impossible, like one code scanned in two distant states within days. Rules catch those with no history and no model.
  4. Signals from three questionsHow was the manufacturer already catching counterfeits? What do we capture on every scan? What would turn an insight into an action? That gave signals on how medicines, batches and single codes move.
  5. My first version was wrongLooking at the codes alone, even normal scans sometimes looked malicious. So I added the IP address and phone device behind every scan, and judged the scanner too. Together they gave a firm answer.
  6. A wrong call costs either wayA missed fake can reach a patient. A false accusation can break a consumer's trust and put a manufacturer's name in the news. So alerts stay with the manufacturer, and each carries its data so their team can check it and tell me if I was right.
  7. How we knew it workedOne manufacturer's known counterfeit codes trained the supervised model and tested the whole. Live, the alerts it acted on turned out to be fraud.
Illustration

Before: verified for the first three scans, over-verified after.

···· 7Q2

4 scans

BeforeOver-verified
With PS AnalytIQNormal
···· 9XD

3,000 scans

BeforeOver-verified
With PS AnalytIQAlert

Why the second code was flagged

  • Elevated verification count
  • Multiple distinct devices
  • Implausible geographic spread
~90%

Less time to surface a suspect code, in testing.

Read the full story

Also at PharmaSecure

  • The person the clients calledKey contact for three of India's five largest pharmaceutical manufacturers, whose serialization projects were delivered on time and without error. They can vouch for it.
  • Packs that cannot be copiedInitiated the non-clonable technology programme and ran its pilot. The pilot is complete, the technology is in its initial launch, and the methodology is under patent review.
  • A system from nothingBuilt the company's inventory management system from scratch. It went live in January 2025, is still in use, and automates operations across six departments, for under a quarter of the original budget.

CSIR-CSIO · Product & Project Associate · Aug 2022 – Aug 2024

Six water quality indices looked at the same Punjab water and disagreed. Then I tried to tell three identical air conditioners apart.

Two years in a government research lab, on water and electricity. I set up the tricity's first real-time water-quality pilot for the Jal Jeevan Mission, and our study of Punjab's groundwater, published in a Springer Nature journal, showed why India needs its own water index. In energy, I built a hybrid model that tells identical machines apart from a single meter. That paper is under review.

My part

  • Set up the tricity's first IoT pilot for real-time water-quality monitoring, which the Mayor of Chandigarh inaugurated in June 2023, and wrote the project reports that took similar projects to state governments.
  • Set up our own lab to test Punjab's groundwater inside a month, and co-authored the paper that came out of it.
  • Built the hybrid NILM model: classification for which air conditioner is on, regression for how much each one draws.
  • Presented the work to our German counterparts at the Indo-German Joint Scientific Workshop in February 2024.
How I thought about it
  1. Start with the questionReal-time monitoring raised a simple question with no simple answer: how do you say, in one number, that water is safe to drink?
  2. Check the measure itselfSix established indices scored the same water and disagreed. That disagreement became the finding: India needs an index built for its own conditions.
  3. Own the bottleneckLabs quoted weeks and a sample spoils in a day, so we set up our own lab and ran every test inside the month.
  4. Build for where it runsOur water framework answers in under a hundredth of a second, fast enough to sit behind a live monitoring station.
  5. Pick the hardest testNILM struggles most with identical machines, so we used three of the same air conditioner, from the same batch, in one room.
  6. Two questions, two modelsWhich unit is running is a classification problem. How much it draws is a regression problem. The hybrid answers both, reading harmonic distortion alongside the usual power readings.
  7. Where it ledLearning what normal looks like and flagging what breaks it became the core of PS AnalytIQ at PharmaSecure.
The NILM lab bench with e-Sense hardware and live dashboard
AC 1AC 2AC 3

e-Sense hardware under test at CSIR-CSIO. The traces are an illustration.

99%

Model accuracy at telling the three apart.

Read the full story

Off the clock

I learned to lead on eight wheels.

  • Gold60th Nationals, 2022, with Chandigarh's team
  • Silver61st Nationals, 2023, as state captain
  • 20+national medals with my team
  • Coachmy students now win national medals
Parth in the Chandigarh jersey at the 61st National Roller Skating Championship
Silver at the Nationals
The Chandigarh senior team with their gold medals and the winner's trophy at the 60th Nationals in Bengaluru
Gold at the Nationals
A 2017 newspaper page headlined: Studies and sports, they show how to straddle two worlds and excel in both
Two worlds, in 2017
The Chandigarh squad after the State Championship win
State Championship
Students with medals and a trophy at the nationals
National medals of their own
Parth with the RHPL 2025 first runner-up cheque
RHPL 2025
The medal wall at home
The wall at home

The rink has its own version of this page. My office walls round off into the boards, and I show you around.

How I think

AI is only worth it where it creates real value. My job is knowing exactly where.

  • Operationalize, don't decorate.AI that ships inside a workflow beats a demo that impresses in a deck.
  • Automate execution, invest judgment.Machines take the manual work; people move up to the calls only they can make.
  • Understand the model to place the bet.Knowing how AI works tells you where it will pay off, and where it won't.

Path

The long way here.

It started with machine learning in a college lab and kept moving toward the people who use the work: avalanche research, AI for utilities, research for the public, a product for drug makers, and now every way a diner orders.

  1. DRDO, Ministry of DefenceML Research Intern

    Where it started. As a computer science student, I built machine-learning models that predicted avalanche-prone terrain around Drass, for DRDO's snow and avalanche lab, reaching 95 percent recall.

  2. Smart Energy WaterQuality Analyst

    My first job was already inside AI: quality for Smart iQ, the AI and machine-learning platform for energy and water utilities, where post-deployment issues fell by 80 percent.

  3. CSIR-CSIOProduct & Project Associate

    Two years of research on water and energy: the tricity's first real-time water-quality pilot, a paper on why India needs its own water quality index, and a model that told identical machines apart with 99% accuracy.

  4. PharmaSecureAssociate Product Manager

    Built PS AnalytIQ, the counterfeit intelligence layer now running live on PharmaSecure's platform, which reduced the time to surface a suspect code by roughly 90 percent in testing.

  5. uEngageAI Product & Experience Lead

    Product lead for customer experience: every way a diner orders, on App, Web, Kiosk and WhatsApp, for 190+ restaurant brands.

Also off the clock

Ninety-plus kids, one classroom at a time.

With the Harnoor Madhar Foundation: from COVID relief drives to getting the children of Jagatpura village into school, and keeping them there.

On the rink I coach too, at the school where I learned. Fifteen or twenty skaters became three hundred who wanted in, and some now have national medals of their own.

Parth reading with children in Jagatpura village
Parth with four of his students wearing their medals

Things I build

The AI on this page is one of them.

How I built the AI you can talk to on this site.

I tested twelve models on the eight questions visitors really ask, including traps and an attempt to hijack it, and kept the one that passed every check.

What the evaluation weighs

  • Groundedness30
  • My emphasis20
  • Answers the question20
  • Way forward15
  • Voice10
  • Consistency5

Tested against 18 hard rules. Some of them:

  • Never invents a role or a result
  • Never shares client names or business numbers
  • Never makes promises on my behalf
  • Can't be talked out of its rules
  • It answers only from the fact sheet. If something is not there, it says so and points to me.
  • It speaks as me, in the first person, and says it is my AI when asked.

Ask it

Read the full case study
  1. Your question
  2. A human checkAn invisible Cloudflare check, warmed up as the chat opens, so it costs a visitor nothing.
  3. Limits and screeningA size limit, a rate limit, and a moderation pass. These now run at the same time rather than one after another.
  4. One grounded modelA single model with a written fact sheet of about 7,900 tokens, generated from this site's own content files.
  5. Receipts and toolsSourced sentences come back tagged with where they live on the page. It can also walk you to a section or take a message.
  6. An answer, with receipts“I don't know. Ask Parth.”
Everything I'm building

You made it to the bottom. Most people don't.

If you're building something interesting, I'd like to hear about it. If you only want to talk roller hockey, that works too.

Or write toparthkhungar33@gmail.com

© Parth Khungar 2026