Most interns spend their co-op term fixing bugs. Ours spent his building agents that improve based on what users actually ask for.
Standard agents do what you tell them. You configure the settings once, and they run. Madhav thought it wasn’t practical enough.
"Usually when you program agents, you set whatever settings you want, and off the agent goes, doing the same thing every time. That isn't actually leveraging AI to its fullest. That's what conventional software does. You program it once, and it just does the same thing every time.”
What he built instead:
“The thing that AI can do, that I was able to enable, is track what the user is asking for, recognize the most common purposes that users are turning to these chatbots for, and then optimize the chatbot settings automatically, so that over time, they'll become better and better at serving customers."
Japanese enterprise documents can run thousands of pages. The same concept might appear across a document written in different ways, because Japanese has different writing systems and the same idea can be expressed using any of them.
Madhav had to build something that could find all those instances, recognize they were the same thing, and link them together fast enough.
"I had to do a lot of trial and error to figure out how to go through so much information extremely quickly so that the user isn't waiting for hours. And it wasn't like there was a theoretical answer in some textbook that was going to solve the problem automatically."
He got something that would manually take hours down to seconds.
Coming into Jurin AI, Madhav was initially wired toward getting the prototype done fast, moving fast... by the end, he was thinking differently.
"Over the course of the internship, I went from putting most of my attention on just getting a prototype done quickly and efficiently, to quality control. Making sure that everything I do has different layers of checks, that I manually test what I'm creating, that I have automated checks to help me catch any issues. Even down to each individual pixel being exactly as specified, instead of two pixels more or less."
At most companies, when something goes wrong, the knowledge stays with whoever experienced it. At Jurin AI there's a shared channel where someone posts what broke, what they tried, what fixed it. Everyone can read it, follow it, and not repeat the same mistake.
"In many prior companies where I've worked, there's a sense of, here's good, here's bad, you have your mission values and all that… but there's no feedback loop to correct things when they go wrong. At Jurin, there are a number of different ways where you can regularly get feedback to change things as they're deviating from the intention. We have a learnings channel where you just post: here's one thing that went wrong, here's how we can not repeat it, and that knowledge is public for everyone.”
This culture came in very helpful when one of the engineers who had to work through a rough feature launch posted what he could do better next time. The next three times Madhav launched a feature, he followed the guide to a T, and it went without a hitch.
“In all the past companies where I've worked, something goes wrong, you figure it out internally for the people it happened to, and the knowledge just stays there siloed.”
On top of the learnings channel, Madhav also discovered another highly functional part of our culture that’s changed how he thought about meetings.
"Our team lead is always encouraging us to just drop off when we're done with updates at our daily syncs. If you do that in most other places, it'd just be: why are you leaving? Are you not committed? Here it just seems natural."
Yes, go ahead and drop off when your part is done. It’s not rude. You do have something better to do. It’s a waste of time and a waste of company resources if you stay for no good reason.
Madhav’s degree sits between hardware and software. With hardware, if you're making a computer chip, you’d spend millions before it’s even produced. Because once it's out there, that's it. You can't push an update to a chip sitting inside a machine somewhere. So every single decision has to be right the first time, and anything that behaves differently than what you designed is a failure.
What he found he liked more, working at Jurin AI, was the opposite of that."The thing I found most interesting to build was self-adapting technology. Things that naturally get better over time. I want to work on technology that can surprise you, it's not like the way you left it yesterday. It's improved itself to better suit your needs. That's what I want to work on."
At the end of the interview, we asked him to describe working here in one sentence:
For an engineer, there's the freedom to explore lots of unknown and new product directions, and then the obligation to refine everything to a 99% quality standard before it ever reaches customers, who are always blissfully unaware of the work that went on.
We hire from the University of Waterloo every term. Look out for Jurin AI on WaterlooWorks if you're interested!
For more information, visit www.jurin.ai.
Jurin AI is building the AI Agent for high-performing enterprises. Our technology automates business communications (email, phone, messaging) and the workflows behind them for a better way to do business.
Within just one year of founding, we have become one of the fastest growing startups globally. With offices in Japan, Korea, and Singapore, we are steering Asia-Pacific towards an agentic future. Trusted by large Japanese enterprises every Tokyo resident relies on and respects, Jurin AI integrates seamlessly into existing tools and systems to execute tasks across 100+ languages, enabling teams to tackle labor shortage, focus on strategic, high-value activities, and finally unlock scale to grow the business by another 10x.
Media Contact
press@jurin.ai
