Gaurub Pandey on Software Services, AI and Talent in Nepal

February 04, 2026
Gaurub Pandey · Bajra Technologies
CEO · Bajra Technologies
AI InterConnect Podcast · Published February 4, 2026

Gaurub Pandey is the CEO of Bajra, a software development company founded in Nepal that builds custom digital solutions for international clients and is developing its own in-house AI systems. He tells Laurent Cochet how a freelance project turned into a company, how his team built a speech recognition system for the Parliament of Nepal and an online exam for thousands of applicants, and why he worries about how junior developers will learn.

From the Bay Area to a kitchen-table startup

Pandey was born in America, moved to Nepal with his family at 10, and went back to the United States for college at Brown University. After a couple of years working in the Bay Area, he returned to Nepal, for no reason other than preferring the social life. He disliked his first job there, quit, and asked friends in the United States for freelance work.

One introduction led to a project too big to handle alone. Having always wanted to start a company, he hired two or three developers and worked out of the kitchen of his apartment. He and his co-founder handled sales themselves for years; after a sales hire did not work out, he became the public face of the company, a role he took on about five or six months before the interview.

Growing a services business on word of mouth

One customer has been with the company since 2015. When loyal clients kept growing their engagements, and five or six customers each had teams of five or six developers, he says the founders saw they could build something bigger and put proper project management in place. Early years with 16 hours a day of load shedding and the 2015 earthquake made the company careful about risk, and taught it to take setbacks in stride.

Bajra does backend, front end and mobile work, with specialties in embedded systems, where it writes software for clients' hardware devices and connected applications, and in sustainability solutions such as waste management and solar data analytics. Its best fit, he says, is a client with some in-house engineering that needs to ramp up quickly in stacks such as Ruby on Rails, Python, React and Angular.

Lead generation has always been the hardest part, he says, and the company has never taken any funding. He believes some of the productivity numbers touted for AI coding agents are a little too high, and that buyers taking a wait and see approach has made lead generation more complicated.

Transcribing the Parliament of Nepal

The Parliament Secretariat, legally required to keep transcriptions current, was about four years behind when it approached Bajra. The team built an editing tool, since 100 % accuracy was never realistic, and started from a speech model released by Meta. An engineer's insight unlocked the project: with only 601 speakers, the model could simply overfit on the parliamentarians.

The team used an Indic version of the model to transcribe past sessions, corrected about 54 hours of data by hand and fine-tuned again. By his account that model reached 95 % for almost all speakers and 99 % for some. The team then found it handled Nepali numbers poorly, such as financial years that span two years, so it recorded three separate hours of audio for numbers alone and fine-tuned further. Separate models handle speaker identification.

Procurement moved slowly, and after massive protests in which, he says, the parliament building was burned down, the project is on hold. The lesson he took is that, at least for now, AI will not remove the human in the loop, and every later project kept room for people.

An AI-assisted exam for thousands of applicants

Bajra runs a traineeship several times a year and, he says, has received up to 4,000 applications for 20 slots. Testing in the office became impossible, so the team built an online exam that checks the mic and camera, forces full screen, flags eye movement or talking with a warning and screenshots, and shuffles answer options. Interviews still follow, so the exam does not have to catch everyone.

To expand the question bank, staff wrote about 300 questions, a chatbot (ChatGPT, as he recalls it) generated about 1,500 variations, and the company reviewed them by hand down to 1,345. He says too many generated questions are still wrong to create them on the fly.

Training juniors and choosing what comes next

The shift he thinks is not talked about enough is how hard it has become to train entry-level developers. He built a lemon orchard app for a friend with Claude in 30 to 40 minutes, but fixing the database choice and config values at deployment relied on his 18 years of software experience. Juniors who lean on AI, he says, often do not know what to do when it cannot fix an issue.

Over the next 12 to 18 months he wants a solid lead generation pipeline and a profitable game studio, run by his co-founder, that makes mobile games. He hopes product revenue reaches a 50-50 split with services. He says Bajra is deliberately not calling itself an AI company yet, because it is hard to stand out in that space and the label can sound less than genuine. Since AI will be part of everything it builds and is a means to an end, he concludes that Bajra does market itself that way, just without saying so.

With fewer constraints he would work on AI in healthcare, using data from cheaper devices to lower costs. His regret is waiting seven years to return to being a technical team first company, and his advice is to listen to your gut, then think it through and trust the people you put in place.

Watch the full conversation

Watch the full conversation above. Gaurub Pandey is on LinkedIn.

Watch on YouTube →

AI InterConnect hosts conversations with founders, investors and operators building with AI. Explore more episodes on the AI InterConnect podcast page.

Back to Blog

AI InterConnect

AI InterConnect connects founders, CEOs, investors and technology leaders through conversations, market intelligence and growth infrastructure.

Explore our podcast, research and GTM systems to identify opportunities, build authority and accelerate business growth.