Why Sikka Built Dental LLM Instead of Using ChatGPT: The Domain-Specific AI Advantage

Why Sikka Built Dental LLM Instead of Using ChatGPT: The Domain-Specific AI Advantage

Most healthcare companies are rushing to wrap ChatGPT in their product. Vijay Sikka built an AI model that beats it.

In a recent episode of Category Visionaries, Vijay Sikka, CEO and Founder of Sikka, a retail healthcare technology platform that’s raised over $30 million, explained why building a domain-specific large language model specifically for dentistry created advantages that general-purpose AI simply cannot match. The result: an AI that outperformed ChatGPT-4.0, Claude, and Gemini on benchmarks while solving the data sovereignty problem that makes healthcare practitioners hesitant to adopt third-party AI tools.

This is the build-versus-buy decision that separates AI theater from AI strategy.

The 30-Year Perspective

When most founders discovered AI in 2023, Vijay was rediscovering it. “My background is 30 years ago I was actually publishing and writing papers on AI and this is when AI kind of died off and then was sexy again,” he explains.

That historical perspective matters profoundly. Vijay witnessed AI’s first winter—when overpromised capabilities led to complete field collapse. He watched funding dry up, researchers scatter, and “artificial intelligence” become a term you avoided in grant applications.

This long view shapes how Sikka approaches AI differently from companies chasing the latest hype cycle. Instead of asking “how do we add AI features?” they asked “what would AI look like if we built it specifically for our market’s unique constraints?”

The answer wasn’t a ChatGPT wrapper. It was Dental LLM.

The Data Sovereignty Problem

General-purpose large language models face a fatal flaw in healthcare: practitioners don’t want patient data leaving their infrastructure.

HIPAA compliance is the obvious concern. But the deeper issue is trust and control. Dentists, veterinarians, and optometrists built their practices on patient relationships. The idea of clinical data flowing to OpenAI’s servers, Anthropic’s infrastructure, or Google’s cloud creates immediate resistance.

This isn’t irrational fear. It’s professional responsibility. When a patient’s dental records, treatment history, and health information leave the practice’s control, practitioners lose the ability to guarantee privacy and security.

Sikka recognized this constraint as opportunity. “It’s an LLM which does not transfer your data out into some mega corporation somewhere,” Vijay explains. “So the dentists and the DSOs and the veterinarians and they all love it because their information stays within the bounds of what they want.”

While competitors integrated third-party AI APIs that required sending data externally, Sikka built an LLM that runs on practitioner infrastructure. Patient data never leaves the practice’s systems. Compliance concerns evaporate. Trust issues resolve.

Data sovereignty isn’t just a regulatory checkbox. It’s a competitive advantage that makes AI adoption possible in markets where practitioners control sensitive information.

Domain Expertise Beats General Knowledge

The decision to build a dental-specific model rather than fine-tuning GPT-4 or Claude proved critical. Sikka created what Vijay describes as “dental LLM which is the first LLM built for dental industry.”

The performance results validated the strategy: “It actually beat on benchmarks. It beat ChatGPT4.0 and Claude by anthropic and Gemini 1.5 by Google.”

How does a smaller, domain-specific model outperform massive general-purpose models trained on trillions of tokens and billions in compute? The answer reveals a fundamental truth about AI in specialized domains.

“That’s because of course it has dental domain. So, you know, I mean, it knows what the dental industry and veterinary industry and others are really looking for. It’s not a general purpose solution,” Vijay explains.

General-purpose models know something about everything. They understand basic dental terminology because they’ve seen it in training data. But they don’t understand the nuanced relationships between procedures, insurance codes, treatment plans, and practice operations that define how dental practices actually function.

Dental LLM knows dental deeply. It understands procedure sequencing, material specifications, insurance implications, and practice workflow patterns. For practitioners in that domain, depth matters infinitely more than breadth.

The Build Decision Framework

When should you build domain-specific AI instead of using GPT-4, Claude, or Gemini? Sikka’s experience reveals the decision criteria:

You have proprietary domain data: Sikka has 150 million patients on their platform, processing a billion transactions daily. This data provides training signal that general-purpose models lack. Without unique data, you’re just fine-tuning on publicly available information.

Data sovereignty is non-negotiable: If your customers absolutely cannot send data to third-party clouds—healthcare, financial services, defense—you need on-premise AI. General-purpose APIs don’t work when data can’t leave customer infrastructure.

Domain complexity creates moats: Dental, veterinary, and optometry practices operate with specialized vocabularies, procedures, and workflows that general models handle poorly. The more specialized the domain, the larger the performance gap between general and domain-specific models.

You have the infrastructure to support it: Building and running custom LLMs requires significant engineering capacity. Sikka spent 10 years building infrastructure before tackling AI. That foundation made domain-specific models feasible.

The performance delta matters to customers: If GPT-4 is “good enough” for your use case, building custom models wastes resources. But if domain-specific performance creates meaningful competitive advantage, building becomes strategic.

Beyond Generation: AI for Business Intelligence

While most companies focus on generative AI for content creation, Sikka deployed AI across their entire platform for insights and analysis. “How do we make sure that we bring the best of benchmarking and real time AI into this industry?” Vijay asks, framing AI as an analysis tool rather than just a generation tool.

With 150 million patients and massive transaction volumes, Sikka uses AI to identify patterns, benchmark performance, and surface insights that individual practices cannot see. A dentist can compare their practice’s performance against thousands of similar practices—anonymized, aggregated, and analyzed through AI models.

This application of AI creates stickiness that chatbots cannot match. Generating text is useful. Understanding your practice’s performance relative to peers is invaluable.

The Life Insurance Expansion

Sikka’s domain-specific AI enabled business model expansion that general-purpose AI couldn’t support. “Oral health and cardiovascular health are connected,” Vijay notes. “I mean, the plaque here is the same plaque that is gathering around the heart.”

Using AI to analyze dental records at scale, Sikka proved a connection: “We have been able to prove, working with some of the largest reinsurers and others, that oral health, if you take care of your oral health, you actually improve your mortality.”

This insight created a new revenue stream. “Our platform and the AI that we use to build our models on mortality are now being used by, you know, around six or seven life insurance companies with consent to do underwriting.”

This business model—using dental data to price life insurance—requires AI that deeply understands dental health indicators and their relationship to overall health outcomes. GPT-4 can’t do this. It requires domain-specific models trained on proprietary healthcare data.

The Manufacturer Research Application

The third AI application extends to manufacturer partnerships. “We are working with, you know, industry manufacturers to hone research and product development, again using benchmarking and clinical data,” Vijay shares.

Dental equipment manufacturers, material suppliers, and pharmaceutical companies need to understand how their products perform in real-world clinical settings. Traditional clinical trials are slow and expensive. Sikka’s platform provides real-world evidence at scale.

AI models analyze how specific materials perform across thousands of procedures, which techniques yield better outcomes, and how different products compare in actual practice settings. This creates research capabilities that manufacturers cannot build internally.

Again, this requires domain-specific AI that understands clinical data structures, procedure coding, and outcome measurement in dental contexts. General-purpose models lack the specialized knowledge to extract meaningful insights from clinical records.

The 2025 AI Strategy

When asked about priorities for 2025, Vijay’s answer was definitive: “Brett, we are AI company. I mean, Sika AI.”

But the focus isn’t on AI features for the sake of having AI. It’s on using AI to transform practice operations through insights that were previously impossible. “How do we make sure that we bring the best of benchmarking and real time AI into this industry?”

With infrastructure connecting to 450 practice management systems, 150 million patients on the platform, and a billion daily transactions, Sikka has the data scale necessary for meaningful AI applications. The question isn’t whether to use AI—it’s how to deploy it in ways that create irreplaceable value.

The Build-Versus-Buy Lessons

Sikka’s experience building Dental LLM offers clear guidance for founders facing the build-versus-buy decision in AI:

Build when you have proprietary data, domain complexity creates performance gaps, data sovereignty matters to customers, you have infrastructure to support it, and the performance delta creates competitive advantage.

Buy when general-purpose models are good enough, you lack unique training data, data can leave your infrastructure, you need fast deployment, or domain complexity is minimal.

Most importantly: don’t build AI for the sake of building AI. Build it when domain-specific performance creates moats that general-purpose models cannot cross.

Sikka built Dental LLM because dentists need AI that understands dentistry, runs on their infrastructure, and delivers insights that only proprietary healthcare data can generate. That’s not AI theater. That’s AI strategy that beats GPT-4 on the metrics that matter.