Building Nepal's AI Future: The Vision Behind Ancoda Labs and Its Mission to Put Nepali Innovation on the Global Map

 

Kathmandu. On freezing nights in Cambridge, after spending long hours researching physical AI inside Harvard University's cutting-edge robotics lab, Suryodaya Bikram Shahi would return to a room without a bed. Unable to afford one after paying rent, he slept on the cold floor.

Those nights gave birth to a question that would change the course of his life.

As artificial intelligence rapidly transformed the world, he wondered whether Nepal’s language, knowledge, and identity would slowly disappear from the digital future. That thought, coupled with a deep sense of emptiness, ultimately convinced him to walk away from the prestige of Harvard, a PhD opportunity, and a lucrative career abroad.

Today, Shahi works from a modest office in Kathmandu’s Maitidevi alongside two longtime friends, determined to build Nepal’s own AI ecosystem and ensure the country becomes a contributor—not merely a consumer—in the global AI revolution.

From Delhi to Harvard

Shahi's journey into artificial intelligence began at Delhi Technological University (DTU) in India, where he studied Computer Engineering.

Years before generative AI and ChatGPT became mainstream, he was already conducting research in medical AI and computer vision. At DTU, he developed AI models capable of detecting foot ulcers and identifying diseases in plant leaves.

He later pursued advanced research at the University of Maryland, one of the world's leading institutions for computer science research. There, he worked under renowned researcher Dr. Cornelia Fermuller on the Human Motion to Audio project, focusing on computer vision, robotics, and human perception. But Shahi had even bigger ambitions.

After months of sending cold emails, facing repeated rejections, and relentlessly pursuing opportunities, he secured a research position at the Harvard Ophthalmology AI and Robotics Lab.

Harvard's Glamour and the Reality Behind It

At Harvard, Shahi became involved in the Meta Aria Glasses AI project, designed to assist visually impaired individuals. The system could understand surrounding environments, recognize scenes, and provide real-time action-based guidance to help users navigate independently. The work also immersed him in cutting-edge fields such as Physical AI and Embodied AI.

Although Harvard appeared glamorous from the outside, life behind the scenes was far from luxurious.

After paying rent, Shahi did not have enough money left to purchase a bed. Every night, he slept on the floor.

His routine became relentless: spending entire days conducting research inside Harvard's laboratories before roaming Cambridge's innovation hubs in the evenings searching for a potential startup co-founder.

Despite working at one of the world's most prestigious research institutions, he increasingly felt something was missing.

As AI advanced at extraordinary speed, Nepal remained almost invisible in conversations shaping its future.

The Real Foundation of AI

While conducting research under Dr. Mengyu Wang at Harvard, Shahi says he arrived at a realization that changed everything.

The future of AI, he concluded, is not driven primarily by better algorithms, but by better data.

According to him, today's most advanced AI systems are increasingly limited not by model architecture but by the lack of diverse, high-quality datasets representing languages, cultures, communities, and knowledge systems across the world.

Countries such as Nepal remain severely underrepresented.

For Shahi, this meant that if Nepal failed to contribute its own language and knowledge, future AI systems would continue to overlook the country.

Determined to change that, he returned to Kathmandu in January 2026, giving up both a PhD opportunity and a well-paying career overseas.

A month later, Ancoda Labs was born.

From Esports Teammates to Startup Co-founders

Shahi says startups are never solo missions. "A startup isn't a Rambo mission. It's a team game."

Instead of assembling strangers, he turned to people he had trusted for years.

One of them was Sushant Raj Dangal, now Ancoda Labs' Chief Operating Officer.

The two met more than a decade ago in Nepal's competitive esports community, where they played games such as Fortnite and Valorant together.

Those years of competitive gaming, they say, taught them valuable lessons about teamwork, resilience, decision-making under pressure, handling losses, and trusting teammates in critical moments.

Dangal later built over six years of professional experience working with technologies including React, Next.js, Node, and Python.

The third co-founder, Arjun Dahal, serves as Chief Technology Officer.

Interested in technology since childhood, Dahal began experimenting with software development, system design, and hacking at just 15 years old. He later worked as a backend engineer at several leading Nepali IT companies as well as AdaniConneX, where he specialized in backend engineering and distributed systems before joining Ancoda Labs.

One distinctive aspect of Ancoda Labs is that all three founders personally write every line of code behind the company's products.

They outsource nothing.

Rather than collaborating across different time zones through online meetings, they work together physically from Kathmandu, seven days a week.

One of Ancoda Labs' biggest innovations addresses a long-standing challenge for Nepali AI. Most large language models are primarily trained around Latin-based languages such as English. Because Nepali uses the Devanagari script, existing AI models often split a single Nepali word into three or more tokens, whereas an English word may require only one.

This dramatically increases inference costs.

According to Ancoda Labs, AI applications operating in Nepali can cost nearly three times more because of inefficient tokenization. To solve this, the company has developed its own Devanagari tokenizer, allowing AI models to process Nepali words naturally within their native script.

Shahi says this innovation reduces AI inference costs for Nepali applications by roughly three times.

The company also claims its tokenizer has outperformed the Nepali tokenizer developed by India's major AI company Sarvam during demonstrations.

The Birth of Danta-Med

While AI appears to be booming globally, Shahi believes the industry is quietly facing a major implementation crisis.

He points to MIT Nanda Initiatives' "The GenAI Divide: State of AI in Business 2025", which found that approximately 95 percent of AI pilot projects fail or never reach commercial deployment.​ Similarly, Scale AI's "The Six Percent Report," published in June 2026, found that only 6 percent of organizations worldwide have deeply integrated AI into their core operations.​ Both studies point to the same conclusion: AI fails not because models are weak, but because organizations lack high-quality data infrastructure.

That philosophy became the foundation of Ancoda Labs.

Rather than building flashy AI demonstrations, the company focuses on creating world-class data infrastructure, enabling organizations to organize, structure, and manage data before applying AI effectively. Its first flagship product is Danta-Med, developed specifically for the healthcare sector. The name "Danta" draws inspiration from Hindu mythology. According to legend, while Sage Ved Vyasa dictated the Mahabharata, Lord Ganesh broke one of his own teeth (danta) and used it as a pen so he could continue writing without interruption.

Ancoda Labs says the product reflects the same philosophy—capturing complex human knowledge accurately and preserving it through structured digital records.

An AI Assistant for Doctors

Danta-Med functions as an ambient AI-powered clinical scribe.

Rather than requiring doctors to manually type notes, it passively listens to conversations between physicians and patients during consultations and automatically generates structured clinical documentation inside the hospital's existing EMR/HMS systems.

The platform includes several advanced capabilities, including:

• Ambient audio capture

• Speaker diarization that distinguishes between doctors and patients

• Automatic SOAP (Subjective, Objective, Assessment, Plan) documentation

• Predictive workflow suggestions, including medication recommendations and laboratory test prompts

According to Ancoda Labs, the system can save doctors approximately two hours of paperwork every day.

It is currently undergoing live pilot deployments at three private hospitals in Kathmandu.

Beyond Healthcare

Healthcare is only one part of the company's broader AI portfolio. Ancoda Labs has also developed AI tools for banks, financial institutions, corporations, and government agencies. ​One such solution is a digital onboarding platform for banks.

Using OCR and facial verification, customers simply photograph a citizenship certificate or passport, after which the system automatically extracts information, verifies identities, and completes KYC and KYB processes. The founders claim this reduces verification time from approximately 20 minutes to just one minute.

The company has also developed Lipi, a Devanagari OCR model designed to digitize government and legal documents. Since many official records in Nepal still exist only as scanned PDFs or images, Lipi can convert approximately 10,000 pages in a single day—a task that would otherwise take months—at an estimated cost of Rs. 5 to Rs. 10 per page.

Another platform, Harmony, serves as a digital twin for organizations by integrating information from HR, finance, sales, marketing, and other departments. Together with Second Brain, it allows executives to interact directly with AI, simulate strategic decisions, and gain organization-wide insights.

Building for the Global AI Economy

Running a deep-tech startup in Nepal has been far from easy. According to Shahi, many local investors still struggle to understand pre-revenue technology startups. He says founders are often pressured to give away as much as 50 percent equity in the early stages while simultaneously being pushed toward quick profits or early IPOs.

Rather than sacrificing ownership, Ancoda Labs has chosen to remain entirely bootstrapped, financing operations through its own resources.

To manage the enormous computational costs associated with AI development, the company has also been accepted into the NVIDIA Inception Program, providing approximately US$100,000 worth of GPU credits and computing infrastructure annually—equivalent to more than Rs. 15.4 million.

The company further sustains itself by using its own data engine to collect, annotate, validate, and prepare high-quality datasets for international technology companies, Y Combinator-backed startups, data marketplaces, and leading AI research laboratories. Beyond generating revenue, these projects strengthen Ancoda Labs' expertise in building the data infrastructure required for next-generation AI systems. As part of this work, team members have personally travelled to monasteries across Kathmandu to locate Dzongkha speakers and collect hours of speech data for multilingual AI projects. Income from such specialized data collection assignments has helped finance the company's growth.

Supporting Nepal's Open-Source LLM

Alongside its commercial work, Ancoda Labs is contributing to Nepal's open-source large language model initiative through a strategic partnership with Himalayan AI. Within the collaboration, the company is helping build critical technical infrastructure, including high-quality datasets, evaluation systems, data pipelines, and foundational data layers necessary for developing a competitive Nepali LLM.

The Next Frontier: Spatial Intelligence

Shahi's ambitions extend well beyond language models.

His long-term vision is to develop the data infrastructure required for Spatial Intelligence—the ability of AI systems to perceive, understand, map, and make decisions within the physical world much like humans. The field represents a continuation of the research he pursued at Harvard.

Future AI systems, he believes, will rely not only on language understanding but also on World Models capable of interpreting physical environments, understanding context, and making intelligent real-world decisions. Such technologies are expected to power future generations of autonomous vehicles, robots, and embodied AI systems.

Ancoda Labs aims to build the datasets, evaluation systems, and data pipelines needed to make that future possible.

From a small office in Maitidevi, Kathmandu, the company is pursuing an ambitious goal: ensuring that the world's most advanced AI systems can understand not only the parts of the internet that already exist, but also the places, languages, and communities that have long remained invisible—including Nepal.

पछिल्लो अध्यावधिक: साउन ९, २०८३ १३:५८

टिप्पणीहरू