Korea’s TANGO Framework is Rewriting Industrial AI Adoption
Since 2021, a subtle but consequential change has been taking place across Korea’s industrial landscape. In factories, hospitals, and shipyards, domain specialists who once relied entirely on software engineers to translate expertise into code are now gaining the ability to develop artificial intelligence applications themselves. This shift has been driven by a practical need rather than academic curiosity. Demand for AI-enabled software has surged across every sector, while the supply of experienced AI engineers has failed to keep pace.
At the centre of this transformation is a software framework developed by the Electronics and Telecommunications Research Institute. Known as TANGO, short for Target Aware No-code Neural Network Generation and Operation framework, the platform has been designed to lower the barrier between domain knowledge and deployable AI. Rather than replacing software developers, it reshapes the workflow so that specialists with limited AI expertise can meaningfully participate in building and operating intelligent systems.
Why Traditional AI Development Has Held Industry Back
In conventional AI application development, responsibilities have been sharply divided. Domain experts handled data collection and data labelling, while software developers and AI specialists took charge of model design, training, deployment, and ongoing operation. That division worked when AI projects were limited in number and scope. It has become a bottleneck as artificial intelligence has expanded into quality inspection, medical imaging, predictive maintenance, and autonomous navigation.
The problem is not a lack of insight on the shop floor or in the clinic. Engineers can spot defects in steel products at a glance. Radiologists can identify tuberculosis from a chest image in seconds. Translating that judgement into an AI model, however, has historically required deep neural network expertise, lengthy development cycles, and complex deployment pipelines. TANGO was conceived to close that gap by automating much of the neural network generation and operational workflow.
Inside the TANGO Framework
TANGO is built as an end-to-end MLOps framework that automatically generates neural networks using a no-code approach and manages their deployment across diverse hardware environments. These include cloud platforms, Kubernetes-based on-premise systems, and on-device environments where latency, security, or connectivity constraints rule out cloud processing.
The framework is deliberately pragmatic. Installation is handled through a simple command, after which users can immediately begin working through a web-based interface. For domain experts unfamiliar with neural network architectures, TANGO abstracts away complexity while still producing models optimised for specific tasks such as object recognition. This design reflects real-world industrial requirements rather than laboratory benchmarks.
From Object Recognition to Generative AI Operations
Early development of TANGO focused on vision-based neural networks, an area of immediate relevance to manufacturing inspection and medical imaging. ETRI developed neural network automation algorithms optimised for object recognition, incorporating feedback from domestic industrial sites to ensure practical usability.
More recently, the framework has expanded into LLMOps, supporting the development and operation of generative AI systems. These tools manage the full lifecycle of large language models, including data preparation, training, tuning, deployment, and operation. This expansion signals a recognition that generative AI is moving rapidly from experimental use cases into industrial and institutional settings where reliability and governance matter.
Open Source as a Strategic Choice
ETRI has released the core TANGO MLOps technology as open source on GitHub, a decision that has shaped its adoption trajectory. By making the source code publicly available, the research team has encouraged experimentation, community contribution, and commercial uptake without locking users into proprietary ecosystems.
To support this growing community, ETRI held its fourth public seminar on 6 November at the Science and Technology Center in Gangnam, Seoul. The event was designed to expand engagement with the TANGO GitHub community and to share practical insights gained through real-world deployments. Over four public seminars, a total of 944 participants from 552 institutions have taken part, underlining the breadth of interest across sectors.
Demonstrated Impact and Commercial Results
The development of TANGO has delivered measurable outcomes beyond academic recognition. Over the course of the project, the research team generated 24 domestic and international patents, published three NeurIPS papers and 13 SCI-indexed journal papers, completed four technology transfers, and achieved approximately KRW 10 billion in commercialisation revenue.
These results reflect a deliberate emphasis on technology readiness rather than theoretical exploration. TANGO has been positioned from the outset as a framework capable of moving from laboratory to factory floor, hospital ward, and operational vessel without excessive reengineering.
Autonomous Navigation and On-Device AI
One of the most prominent commercial adopters of TANGO technology is Avenotics, a company specialising in autonomous maritime navigation solutions. Avenotics was selected for the Public Research Outcome Expansion and Commercialisation Project supported by the Ministry of Science and ICT, a programme designed to bring government-funded research into the market.
Through technology transfer, Avenotics secured TANGO’s on-device deployment and AI performance optimisation technologies. These capabilities are now being used to commercialise on-device AI systems that automatically generate contextual information required by navigators. The company has attracted investment totalling KRW 1.3 billion from Korea Science and Technology Holdings, the Korea Credit Guarantee Fund, and Low Partners, placing its corporate valuation at KRW 9.8 billion.
Manufacturing Inspection and Cloud Deployment
TANGO’s adaptability has also been demonstrated in manufacturing. Weda Co., Ltd., a collaborative research partner, developed an AI service designed for use directly by on-site employees in steel and automotive parts manufacturing. The service supports vision-based exterior inspection of complex geometries, including automotive bumper rolls where conventional inspection systems often struggle.
In parallel, Lablup Inc. collaborated with KT Cloud to launch a deployment optimisation service supporting Rebellion’s domestic AI acceleration engine, ATOM-Max. This work extended into the commercialisation of a GPU cloud rental service, providing flexible access to AI compute resources while maintaining compatibility with TANGO-based workflows.
Clinical Applications and Hospital Deployment
Healthcare represents another critical application domain. Seoul National University Hospital is developing AI technology that uses large-scale chest CT images and diagnostic data to automatically generate diagnostic reports. The system is designed to support cardiopulmonary disease prediction, addressing both diagnostic workload and consistency challenges.
The technology will be demonstrated across four hospitals, including Seoul National University Hospital, Seoul National University Bundang Hospital, Seoul National University Hospital Gangnam Center, and Boramae Hospital. Evaluation will be conducted using real clinical data, providing a rigorous test of performance and reliability in live medical environments.
Generative AI and Industry-Specific LLMOps
To accelerate commercial readiness in generative AI, ETRI is collaborating with Acryl Inc. on the development of LLMOps tools tailored for industry use. The source code for Acryl’s commercial product, Jonathan, is fully open on GitHub, with core algorithms being integrated alongside the establishment of a standard operating environment for industry-specific generative AI applications.
This approach reflects a broader strategy to ensure that generative AI systems are not only powerful but operationally manageable. By embedding lifecycle management into the framework, TANGO aims to reduce the risk and complexity associated with deploying large language models in regulated or safety-critical environments.
Institutional Support and Strategic Vision
The development of TANGO has been supported by the Ministry of Science and ICT and the Institute of Information and Communications Technology Planning and Evaluation through the Automatic Generation of Neural Network Applications and Optimisation of Execution Environment project and the Generative AI Support System Software Framework project. This backing underscores the national importance attached to building domestic capability in AI development tools.
Kim Tae-ho, Software Programme Manager at IITP, described the initiative as follows: “TANGO technology is truly the best open source project in Korea and is contributing greatly to enhancing the competitiveness of the domestic software industry in the field of artificial intelligence development tools.”
Sharing Expertise, Not Just Code
The research team has committed to releasing new versions of the TANGO source code on GitHub every six months. In addition, annual public seminars will continue in the second half of each year, providing a forum not only for sharing code but also for exchanging practical experience gained through deployment and commercialisation.
Jo Chang Sik, Principal Researcher at ETRI, outlined the long-term direction: “We plan to expand the existing Tango project, which utilises vision neural networks, into the field of LLMOps tools that support generative AI. Even in the future, we will share all of our development expertise and provide solutions that can be directly commercialised by the industry through verification.”
Rather than positioning TANGO as a finished product, ETRI has framed it as an evolving ecosystem. By aligning open-source development with industrial validation, the framework offers a model for how publicly funded research can translate into sustained economic and technological impact.
