About
Frontend Engineer at Deel. I dropped out of engineering college in 2014 to teach myself…
Activity
4K followers
-
Nimesh Pancholi shared this"We're here to help you 24x7." The very next line on Zepto's support page: "Zepto does not have an official customer support phone number." I found that page on day six of an email thread about a gift card. I'd bought a ₹10,000 [$105] Amazon gift card on Zepto. It didn't work. So I wrote in. They replied. I wrote again. Six days later, I got a resolution. Six days, for a code that either works or doesn't. A two-minute call would've closed it on day one. Now, I get why the page reads the way it does. The same page warns about fake numbers and spam calls. Scammers love a support number, and Zepto's clearly been burned. And most tickets really are "an item's missing" or "where's my order". A bot should handle those. Nobody needs a human to confirm the coriander didn't show up. I'm not against the bot. I build AI features for a living. So here's the part that gets me. The FAQ screen on that page offers three doors. Issue with a product. Missing items. Order not received. There's no door for "I paid, and what I got doesn't work." That's the one case where a human earns their cost. The money's already gone, and the only thing that can check the code is their system. Automate the questions. Keep a human on the money. It's the line I'd want drawn at Deel too, where what's on the line is someone's salary. "24x7" doesn't mean much if the only thing awake is the bot. If you've designed a support flow: where did you put the handoff from bot to human, and what made you move it?
-
Nimesh Pancholi shared thisI posted this Short a couple of months ago. The Parliament clip in it is older, but the underlying debate has become relevant again. India's domestic investors have become a powerful force in the market and have helped absorb a significant amount of foreign selling. But that doesn't mean we can simply ignore foreign capital. The Nifty has now gone through seven consecutive losing weeks, while FPIs have once again turned heavy sellers in September. Foreign investors can come and go. That's how markets work. But if we want global capital to keep coming back, we also have to ask a harder question: Are we making Indian markets attractive enough for them to stay? Reposting this one because the conversation is still very much alive. #NIFTY50 #indianstockmarket
-
Nimesh Pancholi shared thisThis year India's phone network started showing a caller's name for free, pulled straight from SIM KYC. That name is the product Truecaller built 500 million users on. On Thursday, Truecaller shipped its answer. Connect your LinkedIn, and when you call someone they see your company, your job title and a "Verified on LinkedIn" badge. My first reaction was the lazy one: great, another app that wants my employer. It's opt-in though, and you can disconnect whenever. So the privacy dunk doesn't land. The real reason is CNAP. Calling Name Presentation. The telco shows the name you gave when you bought your SIM. No app, no internet, no crowdsourced contact lists. Operators had a March 2026 deadline and it's live across a big chunk of the country now. A name used to be something Truecaller had and your phone didn't. Now the network hands it out for nothing. So Truecaller needed something the telco can't see. The telco knows who you are. It doesn't know where you work. And for the person picking up, that's useful. Half the scam calls here open with "I'm calling from your bank." A verified workplace makes that lie harder to tell. If a recruiter calls me with that badge, I'm answering. So the feature is defensible. I was wrong to reach for the privacy angle. Here's the part that gets me. Who turns this on? People who call a lot of strangers and need them to pick up. Recruiters, sure. Also every loan agent and insurance seller who has a LinkedIn. Verified tells you who's calling. It doesn't tell you why. A verified sales call is still a sales call. It's just harder to decline. And the announcement doesn't say what happens to the badge the day you leave the company. Plenty of people changed jobs this month without choosing to. You only price what people can't walk away from. Once the name went free, the job title became the thing worth owning. If you've built KYC, caller ID or any "verified" badge: when someone leaves the company, how long does your system keep vouching for them?
-
Nimesh Pancholi shared thisGitHub says 660 pull requests want my review. I counted properly this week. 64 are actually my team's. 11 of them need me this morning. One of mine had been sitting for 5 dayssitting 5 days with no reviewer, and I hadn't noticed. Working that out by hand is 20 minutes every morning: open GitHub, filter, check CI, check approvals, check what's stuck, type a Slack update for the team. Not hard. Just never going to be anyone's priority to fix. Last week I opened Akai, hit record, and did my morning while talking through it. 60 seconds later it proposed a 3-step agent. I named it PR Sweep and set it to run at 10AM. The sorting step doesn't need a model, so I typed "make this step deterministic" and now it runs as plain code. Same result every time, no tokens spent. It scores itself on every run. I tagged a teammate on one step and they can edit it, comment on it, and see every version. I'm a frontend engineer on Deel AI. I build interfaces for AI tools all day. And I still had a 12-click ritual nobody was ever going to write an integration for. Deel used Akai internally first: 90,000 hours of back-office work saved in the first few months. Mine is 20 of those minutes a day. That number is made of a thousand small rituals like this one, not one big automation. What's the 20-minute ritual in your morning that nobody's ever going to build an integration for?
-
Nimesh Pancholi reposted thisNimesh Pancholi reposted thisIntroducing Akai: Deel added $140M ARR in 90 days without hiring any new people, automating ~600 employees' work using akai.run Revenue per employee has 2x'd AND growth is accelerating. We've been automating all our backend tasks. Built >8k agents that do the work of ~600 employees. We built Akai as an internal tool to automate our painfully repetitive operations in Finance, HR, Accounts Payable, and Compliance, etc. We never intended to make this a product. But it had such a dramatic impact on our business that today we are launching it for everyone. How it works: Say you're automating payment reconciliation: 1. Record your screen while manually matching a messy transaction and Akai will capture your screen, voice, server requests 2. Akai will see that you pulled unformatted wire transfer info from an archaic bank portal, put it in some excel sheet, checked NetSuite invoices, payment history, and put a ticket on Zendesk 3. Akai reads between the lines and build a workflow + steps + conditional guardrails. It learns tacit edge cases, like resolving malformed invoice references without you writing a single regex 4. Simply connect NetSuite, your ledger, Zendesk, PSPs, and even legacy bank portals with zero API access 5. Run the workflow and tell it what to adjust in plain English: "strip slashes on wire memos and auto-apply partial payments." It adapts instantly 6. Once it works for you, add 100s of colleagues. Your entire payment ops team forks and extends the workflow for new PSPs, secondary ledgers, or regional settlement rules 7. We automated 85% of our payment reconciliation end to end, eliminating 500+ hours of soul-crushing manual grunt work every single week. Claude Code/Codex can't do this in multiplayer mode. Every person rebuilds the same skill from scratch in their own way. Deel built Akai to: 1. understand backend operations (it had to work for our 7000 person team first) 2. Collaborative across 1000s of employees 3. Self-Learning from millions of runs 4. Optimises cost and gets cheaper every run We're so confident that we're announcing an Automation Guarantee: If our engineers can't automate a thousand of hours of work in your first 30 days, you get a full refund. Book a demo: https://www.akai.run if you're an exec at a company with hundreds of employees Comment ’Akai’ below and you'll get $5000 in free credits + a repo of 100 Akai automations you can start using in your business today.
-
Nimesh Pancholi shared thisThree American CEOs asked the world to slow AI down on Saturday. By Sunday, Beijing had called it fearmongering and Washington had called it a conspiracy. Both against the same essay. Here's the weekend in order, because the order is the story. Saturday. Dario Amodei publishes "We Must Pace the Frontier." Three steps: let independent evaluators inside the labs with employee-like access, get labs and democratic governments to coordinate, then bring China to the table. His warning: within 6 to 12 months, agent swarms could take over large stretches of the internet. Hundreds of billions in damage. Same day, Sam Altman on X: "I agree with Dario that we need to pace the frontier... we will do the same." Same day, Elon Musk: "Dario is right." Sunday. Beijing answers. Foreign ministry: "Fearmongering, confrontation and vicious competition will only disrupt the process of global AI governance." Xi announces a BRICS open-source AI community instead. Trump calls the whole slowdown idea a conspiracy that helps China. Monday. SoftBank down 10%. SK Hynix down 6%. Nvidia down 3%. Anthropic confirms Nasdaq for October. Altman says OpenAI won't list this year. I ship the dialog where a human approves a tool call inside an agent session. So I read the essay the way I'd read a PRD. One line in it has an acceptance criterion. Evaluators with badges and the right to publish. You can check whether that happened. OpenAI matched it within hours. Everything else has no number. No speed limit. No threshold. No penalty for crossing it. "Pacing isn't a pause" is the whole spec. And I want to be fair to the one real line. Letting outsiders read your logs and publish isn't nothing. I've never seen a rival copy a constraint that fast. So the evaluator commitment is real. I was ready to call the whole thing theatre, and that part isn't. But the market didn't sell off on the evaluators. It sold off on a slowdown nobody defined. And the same company that wrote it will price a $2 trillion listing into that undefined slowdown next month. Slow enough to worry investors. Fast enough to list. Both things are true. I haven't squared them yet.
-
Nimesh Pancholi shared thisIn July, Indian investors started 61 lakh new SIPs. The same month, 50 lakh SIPs were stopped. For every 100 people who began, 82 quit. That's AMFI's own number, and it isn't a bad month. It's been above 75 for most of the year. I'm a frontend engineer. A few years ago I started working alongside my dad, who's spent 30 years in finance, and we run a wealth management company together now. So I get to watch what those 82 look like from the other side of the desk. They're not panicking. Most of them are calm. A new phone. A trip. "Markets are high, I'll restart when they fall." "Markets are low, I'll restart when they recover." I used to try to talk them out of it with charts. My dad doesn't. He asks -"Has your salary stopped?" If the answer is no, the SIP doesn't stop either. If the answer is yes, he stops it himself, that day, no lecture. That's the whole rule. It took me a while to see what it's doing. It removes the market from the decision. Every reason people give for stopping a SIP is a forecast. Markets are high. Markets are low. Something better is coming. Every one of those is a guess about next year dressed up as a decision about this month. His question is the only one that isn't a guess. Either the money is coming in or it isn't. The number that gets me isn't the 82. It's what those 82 get instead. A ₹10,000 ($105) SIP paused for a year while you "wait for the dip" isn't a pause. It's 12 instalments that never exist, at exactly the prices you said you were waiting for. I write code for a living. I automate everything I can so I don't have to make the same decision twice. Then I spent years watching 82 out of 100 people make the same decision twice, every month, against themselves. He'd been asking one question the whole time. If you've ever stopped a SIP or an investment you'd promised yourself you wouldn't touch: what was the actual reason that day, and did you ever restart it?
-
Nimesh Pancholi shared thisIndian life insurers paid out 10.3 lakh death claims last year. The average payout was ₹3.4 lakh ($3,550). That's IRDAI's own number for 2024-25. ₹34,995 crore ($3.7 billion) spread across 10,34,455 families. I also work alongside my dad, who's spent over 33 years in the insurance industry, at the wealth management company we run together. So people bring us their policies in a folder, and I've read a lot of them. Your first guess, like mine, is probably that people are buying tiny policies. They aren't. They're paying real premiums. ₹40,000, ₹60,000, ₹1 lakh a year ($420 to $1,050). The premium is just doing a different job. Most of what's in those folders is a savings plan with a small life cover stapled to it. Endowment. Money-back. The plan a relative sold you in your first year of work. And I want to be fair to that plan. It forces you to save. The payout is guaranteed. Maturity is tax-free. For someone who would otherwise spend every rupee, that discipline is worth real money. My dad has watched it work for exactly those people for 33 years. So the product isn't the villain. I was wrong about that for years. The problem is the word "insurance" on the cover. Because a bundled plan usually pays your family 10 to 20 times the annual premium if you die. A pure term plan pays around 1,000 times. Same ₹60,000 ($630) a year. A ₹1 crore ($105,000) term cover costs a healthy 30-year-old roughly ₹9,500 ($100). The other ₹50,000 ($525) goes wherever your savings already go. Bundle the two and you get a 4 to 6% return and a family that receives ₹3.4 lakh. Not because India is poor. Because the premium was busy being an investment. I'm not going to name products. This isn't advice. It's one line. Pull out your policy tonight and find "sum assured." Is that a number your family could live on for ten years? If you've ever surrendered a policy you'd paid into for years: what finally made you do it, and what did the surrender value come to against what you'd put in?
-
Nimesh Pancholi shared thisLast month I ran a speed test at 39,007 feet and got 207 Mbps. I was on holiday, flying Virgin Atlantic VS10 out of New York. Somewhere past Nova Scotia I opened the laptop out of habit, expecting a captive portal, a spinner, and a decent excuse to stop. It just connected. SpaceX's Starlink. Free, no upsell, about ten seconds. So I ran a test. 207.17 down. Then I looked at the ping. 103 ms. I'd spent years thinking plane wifi was slow. It was never slow. It was far. Legacy inflight wifi hangs off satellites parked at 35,786 km. The signal makes that climb twice, so you're starting at roughly 600 ms of round trip before a single useful byte moves. That's the whole reason every call you've ever taken from a plane felt broken. Starlink sits around 550 km up. Same physics, a fraction of the distance. So my 103 ms was mostly not the satellite at all. It was the Atlantic still being under me, and the test server sitting in London. I'd been blaming the wrong number for years. I merged a PR and deployed to Cloud Run from my seat, on a side project I'm building called Ask Gurudev Anything. It went out like it was a Tuesday afternoon. And the number that could actually have stopped me is the one nobody screenshots. Upload: 15.24 Mbps. Everything I did that day was download-shaped and got built somewhere else. If I'd been pushing a container image instead of source, I'd have watched a progress bar for the rest of the ocean. Bandwidth is what gets advertised. Latency decides whether you can work. Upload decides whether you can ship. I was on holiday. I opened the laptop anyway, and I'd do it again. I haven't worked out yet whether that's a good thing. If you build software: what's the number you spent a week blaming before you found the real one? Bandwidth, memory, a query that turned out not to be slow. Tell me the specific chase.
Nimesh Pancholi commented on a post
1w
AKAI has been game changing for us here at Deel
Nimesh Pancholi replied to a comment
1mo
Tamal Das Good question, and the timing is better than you'd think, because Indian fintech has already been pushed off this exact pattern.
RBI's authentication directions came into force on 1 April 2026. Every digital payment transaction now needs two factors with at least one dynamic, and the named alternatives are passkeys on FIDO, device-bound public key cryptography, and biometrics. The direction of travel is away from SMS OTP entirely.
Look at what those have in common: the credential is never a string a human can say out loud. You can't repeat a fingerprint down a phone line. You can't forward a device-bound key to a caller.
Swiggy's login isn't a payment transaction, so none of that governs it. But it still ends in six digits, and six digits can always be spoken.
.
-
Nimesh Pancholi liked thisNimesh Pancholi liked thisWe built our own npm registry with AI. It took a few endpoints, not a moonshot 🛠️ The trigger was a familiar fear. Deel has 200+ repositories. A popular npm package gets hijacked, and someone has to answer: are we running the bad version, and where? 😬 Answering that by scanning lockfiles is brutal. Three package managers, multiple lockfile formats, side branches that never show up on master. And even a perfect scan is too late, because the postinstall script already ran on someone's laptop. Scanning fights the problem at the wrong layer. Every install goes through a registry, so that's where the control belongs. 🎯 The npm registry protocol is just HTTP: metadata, tarball, publish, auth. npm, pnpm, and yarn all speak it. So we built a caching proxy behind our VPN and pointed every developer and CI job at it. What that gave us: 🛡️ Malicious versions blocked before install, not found after 📏 Release-age rules and scanning enforced server-side, for every package manager 🔍 "Are we affected?" becomes a database query, not a fleet-wide grep 📦 Our own S3 copy of every dependency, so upstream outages don't break builds Prevention instead of detection. Once you own the choke point, a whole class of supply chain problems gets tractable. ✅ Full write-up on the Deel blog: https://lnkd.in/dsjVQUnJThe npm registry protocol is a handful of HTTP endpoints. So Deel built its own.The npm registry protocol is a handful of HTTP endpoints. So Deel built its own.
-
Nimesh Pancholi liked thisNimesh Pancholi liked thisOver the last few years, my perspective on finance has changed. Earlier, I thought being good at finance was mainly about getting the numbers right. Today, I think it’s also about knowing why the numbers are changing, what doesn’t look right, and what can be done better. With AI and automation taking over more repetitive work, finance professionals have an opportunity to spend less time just processing data and more time understanding it. My CPA journey reinforced one thing for me: Strong accounting knowledge + technology + curiosity = a much more valuable finance professional. The numbers still matter. But knowing the story behind them matters even more.
-
Nimesh Pancholi liked thisNimesh Pancholi liked thisYou're about to say "like I said..." but you stop yourself. You address something and not a minute later, the person you're talking to asks a question about the exact thing you just addressed. But it's not a follow-up and it's definitely not a clarification. It's as if the subject never came up. I'm always stuck here. Do I say "like I said..." and show that I'm frustrated, or do I fake enthusiasm and say "great question" and take it as a chance to make my answer better? Or do I just answer with zero emotion and move on? It's this choice between looking irritated, looking fake, or looking resigned.
-
Nimesh Pancholi liked thisNimesh Pancholi liked thisWhen someone's salary doesn't land, every hour matters. Failed payments are usually small things. A digit off in an account number, a closed account, a name that doesn't match the bank's records. But for the person waiting on their pay, it isn't small at all. Until recently, fixing these meant a lot of manual work: scanning failures country by country, working out what the bank actually rejected, then writing to each person one by one. So I built an agent on AKAI, Deel's own agentic automation platform, to handle it. It now runs on its own across 100+ countries. It finds the failed salary payments, reads why the bank rejected them, and emails the affected worker in plain language with exactly what to correct. Anything that genuinely needs a person gets routed to the right owner. No queue. No waiting for someone to notice. That's what makes AKAI different from most AI I've worked with. It doesn't summarize the work, it does the work, learns from its mistakes, constantly evolves and gives you critical insights. And the hours it gives back go into the things only a person can do. AKAI is now available to customers, not just Deel teams. If your team is stuck running repetitive checks across systems that don't talk to each other, it's worth a look. https://www.akai.run
-
Nimesh Pancholi liked thisNimesh Pancholi liked thisI automated a task by doing it once, slightly badly, while something watched. No flow builder. No trigger config. No screen where you map field A to field B. I hit record, narrated what I was doing the way I'd explain it to a new hire, and finished the task. What came back was a working workflow. And underneath it, a set of principles Akai had written for itself. Rules I never said out loud, that it inferred from watching where I slowed down. That's the part that changed my mind. Every automation tool I've used asks you to translate your judgment into its syntax. Judgment is exactly the thing that doesn't survive the translation. This one watches you exercise it instead. It also keeps every run in memory. It hits an edge case specific to our setup, handles it, remembers it. There's an eval view showing adherence and efficiency run over run, so you can watch it get better rather than take it on faith. Nobody involved needed to be technical. At Deel this added up to 90,000 hours of back-office work removed in a few months, and in a single month, the equivalent of 500 full-time employees' hours, at a company of 7,000 people. Akai is live today. For companies over 100 people, Deel is guaranteeing 1,000 hours saved in the first 30 days or a full refund, which makes it a fairly cheap bet to take. What's the thing in your Monday morning you've never bothered automating? Put it below and I'll tell you honestly whether this handles it.
Experience & Education
Licenses & Certifications
Volunteer Experience
-
Life Skills Coordinator
The Art of Living
- Present 12 years 7 months
Social Services
I volunteered in assisting various courses organized by The Art of Living Foundation . These courses aim to relieve individual stress, social problems, and violence. It teaches people Yoga, meditation and gives them practical life knowledge. I’m myself a regular practitioner of Yoga and meditation.
Courses
-
Android For Beginners
-
-
How Computers Work
-
-
Learn Git
-
-
Learn the Command Line
-
Projects
-
Neighbourhood Map
-
See projectAllow users to search for places near them and provide various details related to those places such as reviews, contact number, opening hours, etc. Build using Knockout JS (MVC framework) and Google Maps APIs along with leveraging service workers. This project was built from scratch.
-
Restaurant Reviews
-
See projectMobile Web Specialist Nanodegree final project. Progressive Web app with support for accessibility. The job was to incrementally convert a static webpage to a mobile-ready offline web application. Users can add comments while being offline.
Languages
-
English
-
-
Hindi
-
Recommendations received
-
LinkedIn User
1 person has recommended Nimesh
Join now to viewExplore more posts
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More