Insight

Saltlux, the Pride of K-AI, Breaks Through AI Hallucinations and Data Leaks

TECHWORLD ·

Founded in 1981, Saltlux has built high barriers to entry in the domestic B2B and B2G markets with its proprietary agentic AI technology. [Photo: Saltlux]

✦ AI Summary

As the pace of global AI development accelerates, concerns about corporate security and AI hallucinations are spreading.

Accordingly, demand is rising for preventing the external exposure of sensitive in-house data, as well as for on-premises and sovereign AI deployments.

Based on its proprietary large language model Luxia and agentic AI, Saltlux is targeting high-security markets such as finance, the public sector, and manufacturing, as well as B2B and B2G markets.

As the pace of global AI development accelerates, concerns about corporate security and AI hallucinations are spreading as well. Accordingly, demand is expanding to prevent the external exposure of sensitive in-house data, and demand for on-premises deployments and sovereign AI deployments is also surging.

In this environment, the need to reduce dependence on global large language models (LLM) is coming to the fore. At the same time, agentic AI technology that connects directly with in-house legacy systems is emerging as a differentiating factor, and companies with such technology are building technical barriers to entry in the domestic B2B and domestic B2G markets.

Saltlux (Saltlux) is a first-generation Korean AI specialist that is leading the enterprise generative AI ecosystem based on its proprietary large language model, Luxia. This article focuses on Saltlux’s leap from a first-generation AI company to a sovereign LLM and agentic AI heavyweight.

Saltlux was founded in 1981. In the 2000s, it transformed itself into an AI and big data solutions company, and it is a representative fabless AI specialist that laid the foundation for Korea’s AI industry. It also possesses core technologies in conversational AI, natural language processing (NLP), and cognitive-model-based augmented analytics, and it has supplied domestic AI platforms to institutions with high-security requirements in finance, the public sector, and manufacturing.

Saltlux listed on the KOSDAQ market in 2020. Then, in March 2023, it independently developed and unveiled its Luxia lineup, a proprietary large language model. The unveiling of Luxia changed the landscape of Korea’s AI software industry, which had been reliant on big tech API.

Recently, Saltlux has been focusing on Agentic AI rather than response-oriented AI. Agentic AI is characterized by integration with in-house legacy systems, autonomous judgment, and direct execution of tasks. Saltlux has also recently been focusing on building on-device and on-premises platforms, and it is making a leap as a global sovereign AI heavyweight.

Saltlux presents its proprietary large language model Luxia, tailored to internal corporate knowledge assets, and its advanced retrieval-augmented generation technology, Agentic RAG, as core competitive strengths. Saltlux is putting forward Luxia, optimized for internal enterprise knowledge assets, and advanced Agentic RAG as its key weapons.

This stands in contrast to large cloud-based LLMs, which carry the risk of sensitive data leakage and the limitation of hallucinations. Hallucination refers to the problem of answering with irrelevant false information as if it were true. Luxia has a function that converts in-house documents into a sophisticated ontology-based knowledge base. The mechanism of Agentic RAG deeply analyzes the user’s intent in advance, searches scattered internal materials, and generates answers based on source-backed reference documents by referring to them in real time. In this process, the answer includes cited sources, dramatically reducing the hallucination rate. In addition, not a single byte of data leaks outside the corporate network, creating a fully secure environment.

This structure applies to the finance, healthcare, and public-sector markets. These markets fundamentally require the highest levels of security and accuracy. Saltlux explains that in these markets, Luxia and Agentic RAG have become irreplaceable core mechanisms.

In competition with low-cost Chinese AI or software based on general-purpose API, Saltlux highlights the hardware-integrated AI appliance Luxia ON and its agent integration architecture based on MCP as core technologies.

Luxia ON is a single-box integrated appliance composed of high-performance GPU hardware, the Luxia model, and the platform operating environment. The appliance launches a secure in-house AI environment the moment power is applied, completing an on-premises computing structure.

Using the MCP (Model Context Protocol) tool ecosystem, Saltlux combines AI functions with in-house legacy systems beyond simple question-and-answer interactions. As a result, it connects with ERP, CRM, and databases so that agents can check unprocessed purchase requests, draft them, and directly carry out follow-up tasks based on customer consultation records.

Saltlux’s Luxia ON blocks network bottlenecks and, based on performance that processes dozens of complex workflows per second with no delay, secures software-hardware convergence efficiency and is presented as the foundation of platform competitiveness.

At the same time, as sovereign AI demand rises amid global security and data sovereignty issues, and as demand grows for the establishment of proprietary AI systems by governments and companies, Saltlux is being assessed as having entered a sharp inflection point in momentum for winning orders for its proprietary platform.

If the existing revenue structure had been concentrated on one-time license and deployment revenue, it is now expanding its revenue model through sales of the Luxia ON appliance and the accumulation of agent subscription licenses, and it is shifting toward the full-fledged rollout of a high-margin B2B business model.

At the same time, global expansion of the B2C-only AI search and analytics platform Goober and the acceleration of ecosystem building for the next-generation agent builder through annual AI conferences are under way, and as this overlaps with the monopolization of public-sector AI transformation (AX) contracts and entry onto the path to supplying turnkey solutions to global frontline companies, corporate value revaluation is proceeding, leading to expectations that an earnings turnaround will materialize.

Saltlux’s challenge is clear. To leap into a global AI software company, it must move beyond a situation in which a technology gap remains versus global big-tech large-scale models and competition in parameter expansion continues. To that end, it needs to pursue the advancement of its proprietary model and make sustained R&D investment.

At the same time, in the process of pursuing entry into global markets, Saltlux faces labor costs, GPU infrastructure buildout costs, and marketing fixed costs. Accordingly, it needs visible mass-production-style turnkey revenue to offset those fixed costs.

The key to solving this lies in proving a high-profit leverage structure by achieving consecutive wins in Luxia ON license contracts and consecutive wins in Agentic AI license contracts for the B2B enterprise market. This is expected to be the key to enhancing long-term corporate value.

A senior researcher in the AI division said that the more dependence on big-tech cloud API intensifies, the more data security threats and hallucinations come into focus. He analyzed that the value of on-premises LLM and agentic AI that operate entirely within the internal network will expand dramatically, and that Saltlux has its proprietary large language model Luxia and also has a fully integrated hardware solution. He added that Saltlux is expected to directly benefit from the sovereign AI and enterprise AX markets.

Source: TECHWORLD · Kim Yong-su
Original: https://www.epnc.co.kr/news/articleView.html?idxno=407376

References

This article was produced with the help of an automated content generation algorithm.


Source: TECHWORLD

View original

This article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.