"Even with long hours at an early-stage startup, I don't feel like I'm working too much because the platform, the domain, and the team are all interesting. Finding that fit makes all the difference."
Poorvith Gowda had never built an AI product before. 6 months in, he's the full-stack engineer scaling an AI market research platform already used by enterprise clients.
In January, SkillsRobin introduced Poorvith Gowda to Salem at Muja. Muja is an AI market research platform with paying enterprise customers and 4 people. Poorvith became their full-time full-stack engineer.
His learnings since joining the Muja team:
Let's get into it.
Tell us a bit about yourself.
I've been working as a software engineer for close to 5 years. I started out at Accenture, a big enterprise tech company. Thousands of employees, and maybe 50 people on a single project team.
Then during my masters I moved to Australia from India and joined a startup that came out of one of my professor's PhD research. Their research got a bit of funding and they wanted to turn it into a tool, so I worked there for almost 3 years while studying. We built 4 or 5 tools focused on sustainability, which software architects and construction companies could use to design better, more sustainable buildings.
Now I'm a full-time software engineer at Muja. This one is different from anything I've worked on before, because it's an AI startup in the market research space.
What is Muja trying to do?
Before AI, and still now, companies do market research by running focus groups and long research projects. That means hiring a market research firm, often keeping them on retainer, and waiting weeks for results.
With Muja you can do that very quickly and for a fraction of the cost.
Recently our enterprise customers have been running proof-of-concepts, comparing our results against research done the traditional way with focus groups. Our results have shown 89-94% alignment with what their traditional focus group and survey research found, on work that takes them weeks. For Muja, it takes 30 minutes to an hour.
A number of those POCs have converted into paid work with enterprise clients across pharmaceuticals, quick service restaurants, consumer appliances and higher education. Even before I started, there were already paying enterprise clients. That's how the company kept three full-time people and one part-timer employed without ever raising money.
What does your role actually involve?
The platform was already live and serving enterprise customers when I joined, so a big part of my job is keeping that stable. On top of that, it's everything from designing a simple front-end component to building an end-to-end feature from scratch, plus maintaining the infrastructure.
Because we already have paid customers, more risk and responsibility comes with that. We have to make sure those customers can use the platform 24/7 with close to zero downtime.
The way we work on AI features is that our AI engineer builds a prototype first, just to see whether the thing is possible at all. Then I take that and turn it into a real end-to-end feature. So I'm wearing different hats depending on the time of day.
What attracted you to the opportunity?
A few things, but the biggest one was the small team.
When I was at a big enterprise company, the role was narrow. Day in, day out it was the same domain, the same work, the same everything. That becomes repetitive after 6 months or a year depending on the project. Here I see new challenges and new problems every day.
The other thing is speed. In big companies, if you want to try something new you have to ask your senior, who has to ask their senior, and by the time you get an answer you might have already decided not to bother. Here I can think of something, implement it, and talk to my boss. There's one person I report to. If they're happy and everyone's happy, it goes to production.
What's different about working on an AI product?
There are good things and there are risky things.
Whenever you're integrating AI, frontier model or open source, there's always a risk. You've probably heard the term jailbreaking, where people try to break a product and get the model to do things it isn't supposed to do. Even frontier labs worth billions of dollars deal with this every time they ship.
But the flip side is that you can build things that simply weren't possible before.
One feature we built is automatic PowerPoint generation. Muja is a market research platform, so before I joined, the end result was a written report, pages and pages of it. Turning that into a presentation was manual work. Now we take the report, synthesise it, and produce the same information as concise slides that a marketing team can present to their leadership.
That was not technically possible before AI. Even with AI it can be very hard to get right and if the output isn’t good enough, people would end up doing it manually anyway. Getting that right was a real differentiator.
What mistakes would even an experienced developer make the first time when building AI products?
The main one is that AI is non-deterministic. Run the same thing twice and you won't get the same output. Ask the exact same question again and the answer will be different. Having a proper set of guardrails helps enormously.
So you need to be very clear about what output your feature actually needs, because even with the best prompt engineering and the best constraints, it can still break. Security also gets more complicated, which is why guardrails are designed in from the start rather than bolted on afterwards. You have to assume unexpected inputs and build for them.
What was the biggest mistake you made when you got started?
I'd been using AI tools to help me develop products for a while, but I'd never built AI-integrated features. So the thing I'd forgotten was: every API call costs you money.
If you make a mistake in your code that ends up calling the endpoint again and again, that can look completely normal from the code side. But every call costs. I had to make sure every single LLM call being made was actually necessary, not just calling them because they're there.
The second one was testing. Because the model is non-deterministic, you can test 10 times or 100 times and get different responses. You have to work out when to stop testing and when to stop refining the prompt, but generally you need to test a lot more than you think initially.
Who's on the team?
There's Salem – the founder, me, an AI engineer, and a data analyst. That's the whole company.
What's the culture like?
Everyone is available throughout the day. I've had questions at 8pm, 9pm and someone’s been able to ask them. That's not the case in enterprise companies.
Even though it's only been 6 months, I can talk openly with everyone about whatever's going on. Every Monday when we meet we talk about what we did on the weekend. That sometimes runs 30 or 40 minutes. And when I made mistakes early on, the team was welcoming and forgiving about it.
How many hours are you working?
Right now, long hours, like any early-stage startup.
But I haven't felt like I'm working too much, because the platform, the domain and the team are all interesting. Personally I don't feel like I'm putting in a lot of hours.
What's on the roadmap?
The biggest thing is taking the product public. Right now, it’s closed access to a select number of customers. There's no public sign-up. Moving to that is the big item on the roadmap.
For everyone on the team that's a serious piece of work to make sure we can handle the increased load. As the primary full-stack engineer, I need to make sure the infrastructure is scalable and there's no downtime. Our AI engineer is focused on keeping the AI layer robust as usage grows. And our data analyst has to make sure all the data is properly compiled and analysed.
What was the process for finding a job on SkillsRobin like?
The reason I gave SkillsRobin a shot was that it looked like a different kind of platform for applying to jobs. Within the first 20 or 30 minutes I was like, oh, this is an amazing platform.
Two reasons. First, all the new and up-and-coming startups were listed there. Second, I could show interest and have the employer reach out to me, rather than spending hours making an application and tailoring a resume to be pinpoint-perfect, then repeating that for every single company. That's where I'd wasted time on other platforms.
And you reaching out personally gave it a personal touch.
Once the interview process started, Salem was very good. I still remember the first call, I could talk to him comfortably. He made sure the whole process felt comfortable, because in a comfortable setting the person you're hiring will be much more open and less nervous.
Why do you think Salem picked you?
I think it was my communication.
After each round there were take-home exercises, system design, that sort of thing. The next round would be about explaining what I'd done. His feedback was that my communication was very clear and I could break the problem down and explain it properly.
Salem isn't a technical person. He has a good understanding of software engineering and AI, but he hasn't sat down and written code. So I knew I had to explain things in a way that was accurate and interesting. If you explain a technical concept to a non-technical person and make it completely boring, full of technical jargon, they'll stop listening 2 or 5 minutes in.
So it's about maintaining the balance. Keeping it accurate while keeping it simple enough that we both understand what I'm saying.
"Interviews are strange things. You spend an hour trying to discover how someone thinks, and most candidates spend the same hour trying to hide it. Poorvith did the opposite. He showed me his working, wrong turns included and explained it so clearly that I understood problems where I had no business understanding. That is a rarer skill than the code itself, and SkillsRobin found him for us."
– Salem Lassoued, Founder, Muja.ai
Where should someone start?
Even in an AI world, the same software engineering principles still exist. AI is an add-on on top of that.
I started out as a back-end developer, not a full-stack one. I've seen companies, including AI companies, with specific roles for back-end, front-end, DevOps, cloud infrastructure. But if you want to join an AI startup, having a certain amount of knowledge across all of it is very beneficial, even if you're not an expert in every area. On a small team you'll be doing front-end work one day and debugging a cloud deployment issue the next.
We have AI coding tools now, so it's easier to get started on unfamiliar ground. But you still need enough underlying knowledge to ask the right questions while you're debugging.
The other thing: build full-stack apps and try to get real users on them. When you're showing your work to a startup, they'll appreciate actual users far more than a university project. It's getting very competitive. There are a lot of people coming into software and AI now, including people without a CS degree.
And communication again. In an interview you have 45 minutes to an hour. You don't get days to prove what you know. Even if you're an expert, explain concepts simply rather than relying on jargon.
You came to Australia as an international student. How did you navigate that?
That was a bit of a transition. I grew up in India, worked in India, studied in India, so I didn't know how to approach it initially.
Then at university I got onto a research project that got commercialised, and I ended up working on it for 3 years. That was the turning point. It gave me the confidence that my experience from another country was valid and useful here, and would be considered. Yes I was at Accenture in India, but it's still a software company, it's still the same software world.
Personally I don't think it should matter whether you've worked in Australia or anywhere else. It's the same software, the same coding. But I have seen it matters to some people. When I was applying to companies before I moved to Muja, my international experience wasn't always taken into account.
So there are both sides of it. But my advice is to keep trying. You'll find a company where the founder or the manager validates your experience and values it. Salem did.
