[Digital Today reporter Chi-gyu Hwang] "We will compete with a K-Palantir model. We are focusing on ontology development and plan to use open-source large language models (LLMs). We believe this can give us an edge in the enterprise AX market."

Mobigen, known for its big data platform business, is speeding up its push into the enterprise AI market with an ontology-first strategy like that of U.S. software company Palantir.

Um Tae-deok (엄태덕), Mobigen's CTO, said in a recent interview with reporters that for AI like ChatGPT to be used in the corporate market it must be connected to organisational data. Ontology will play a key role, he said. He also stressed a vision of becoming a Korean Palantir by positioning Mobigen as a cheaper alternative to Palantir.

Mobigen has been stepping up an ontology-centric AI strategy since last year. It decided it was better to focus on technology that helps companies use AI well, given that competition in LLMs is already intense and requires substantial capital. Benchmarking Palantir, which is expanding its stake in the enterprise AI market with ontology, Mobigen released its Grapio AI solution platform last year. It has also launched Grapio 2.0 after stabilising version 1.0. It will introduce a 3.0 version next year.

At a high level, ontology can be summarised as a methodology that connects data scattered across an organisation through relationships. It has been around for some time, but has not spread in the market as much as expected. The concept sounds plausible and useful, but it proved difficult to use in practice and involved many obstacles.

The concept of ontology itself is not difficult, but building proper tools to implement it is not an easy task. From a user's perspective, buying an ontology solution does not immediately produce results. Companies must build pipelines to connect source data and also create agents. That requires significant time and cost.

Palantir is a case that lowered barriers to make ontology relatively easy to use in practice and rose into the ranks of leading technology companies. Its weight in the enterprise market has grown further with the emergence of LLMs represented by ChatGPT. Palantir provided a data relationship network that AI can understand. That led to Palantir having a presence in the enterprise AI market comparable to that of LLM companies.

Mobigen, broadly speaking, is using tactics similar to Palantir. Grapio also turns the abstract concept of ontology into a usable solution for real-world work.

Um said ontology technology is closer to a new work system than to RAG (Retrieval-Augmented Generation), which is used simply for retrieval. Unlike RAG, which retrieves information in corporate databases based on meaning, ontology also allows data input like existing business systems. When data is entered into a Palantir platform, it can be reflected in other connected systems as well. That is why Palantir calls its solution a data OS, Um said.

Like ontology, relational databases are also relationship-based. But existing relational databases alone have limits in connecting data through relationships across an enterprise. "Existing relational databases are table-based, and the relationships are siloed," Um said. "In contrast, ontology platforms like Palantir can connect data around relationships across all systems and keep extending those relationships."

Knowledge graphs and ontology are similar in concept, but differ in many details. Knowledge graphs originated in search, while ontology emerged to represent human thinking and supports more complex systems than knowledge graphs through reasoning, Um said.

Like Palantir, Mobigen aims to lower the entry barriers companies face in implementing ontology through Grapio. It is focusing on becoming a cheaper alternative to Palantir.

Um said there are already many customers in South Korea looking for ontology. Many bid requests for proposals (RFPs) also include ontology, he said. "We will focus on targeting vertical markets such as defence, EPC (Engineering, Procurement, Construction), the public sector and telecommunications," he said. "We are also nurturing an FDE organisation," he added, expressing confidence.

Grapio 2.0, which Mobigen recently unveiled, has a four-layer structure: data, ontology, apps, and governance and collaboration. The core is the ontology layer, which connects scattered data and embeds context. The app layer supports linking to LLMs based on ontology and allows the registration of agents.

Mobigen also highlights that, aside from price, it places relatively greater emphasis on covering unstructured data as a point of differentiation from Palantir.

"Palantir started by connecting structured data with ontology and is expanding to unstructured data, but Mobigen places relatively more weight on unstructured data," Um said. "In Grapio 3.0, we will further increase the share of unstructured data," he said.

Um's emphasis on unstructured data reflects his view that as AI spreads, the IT environment will shift faster from deterministic systems, where output is determined by input, to probabilistic systems. "Recent AI characteristics are language-based and probabilistic," he said. "The key is how to mix deterministic and probabilistic elements. Recent technology trends show the weight of probabilistic elements will keep increasing," he added.