Frequently asked questions

TWIGFARM is a content AI company that helps teams understand and use video, documents, images, and audio. Our solutions cover content search, translation and dubbing, AI video production, brand visibility analysis, application review, training datasets, and knowledge graphs. Founded in 2016 with a focus on machine translation research, we have applied our technology to projects in media, the public sector, and enterprise.

Copy link to this question

TWIGFARM was founded in 2016 and began neural machine translation research that year. We developed a global collaboration platform in 2018, began multimodal data research in 2021, and launched LETR WORKS in 2023. We established our Singapore entity and Gwangju AI Data Center in 2023 and participated in NIA AI training-data projects for six consecutive years from 2020 to 2025.

Copy link to this question

Our headquarters is on the fourth floor of Penta Tower, 9 Seoun-ro 26-gil, Seocho-gu, Seoul. We also operate a Corporate R&D Center in Gangdong-gu, Seoul, the Gwangju AI Data Center, and a Singapore entity. Addresses and maps are available in the locations section of the Company page.

Copy link to this question

We have delivered projects across broadcasting and media, entertainment and webtoons, education and publishing, public institutions, and enterprise and global services. The Company page lists clients and partners including SBS, CJ ENM, Kidari Studio, Woongjin Thinkbig, and NIA. The scope differs by client; the AI Localization page also describes projects and deliverables by content type.

Copy link to this question

We offer eight solutions: LETR WORKS for content search and management; AI Signal for brand visibility in AI answers; Preview Note for footage review; AI Localization for professional translation and dubbing; Data Intelligence for AI training datasets; AI Content Production for recurring video; Litmus for application review; and LETR K-Graph for connected facts, relationships, and sources.

Copy link to this question

You can manage video, images, documents, and audio in one place and search for relevant content or scenes. Subtitling, translation, and dubbing continue from the same original, with derived files managed alongside it. You can also share an online screening room with approved buyers and review viewing activity. Video search indexing is enabled selectively for the assets that need it.

Copy link to this question

The current service name is LETR WORKS. We plan to rename it LETR AI, but the change has not taken place yet. These are the current and planned names of the same service.

Copy link to this question

LETR WORKS is a platform for teams to manage content and run search, translation, and dubbing workflows. The professional service adds translators and reviewers who check AI drafts and deliver files in the agreed format. The appropriate setup depends on whether your team has its own translation and review staff or needs that work delivered for you.

Copy link to this question

AI Signal examines how often AI answers mention a brand or work, how they describe it, and which sources they cite. It compares Korean and English answers from three AI models to identify name confusion, omissions, and incorrect descriptions. Eight internal brand checks and content checks covering metadata, demand, viewing paths, and reach help prioritize improvements.

Copy link to this question

Results include the questions and responses, mentions and citations by model, incorrect information, and prioritized improvements. After updating your pages or content information, you can repeat the measurement under the same question, model, and language conditions. AI answers vary over time, and diagnostic scores do not guarantee revenue growth or placement in a particular answer.

Copy link to this question

Preview Note organizes speakers, dialogue, and scene descriptions with timecodes, reducing manual transcription and footage review. Scene headings and speaker information help production teams find useful sections and prepare for editing. The output is a draft for review: names, overlapping speech, and descriptions of unusual objects should be checked against the original footage.

Copy link to this question

Provide the original footage and, optionally, a production brief with program, cast, and location information. Names from the brief help identify people and terminology. You receive a DOCX file in the broadcaster’s format and matching Markdown that your team can edit and share. Quality and processing time are checked using your footage and document template.

Copy link to this question

LETR WORKS supports translation across 16 languages, with the available language pairs checked for each content type and function. Formats include MP4, MOV, and MXF video; SRT, VTT, and TTML subtitles; PSD webtoons; and PDF, DOCX, PPTX, XLSX, and HWPX documents. Processing and delivery formats depend on the source layout, editability, and target language.

Copy link to this question

Video projects deliver translated subtitles, dubbed audio, and video in agreed formats. Webtoon projects deliver editable PSD files and PNG previews, while document projects use agreed document formats. Specialists compare AI drafts with the original and check wording, context, timing, and layout. Existing translation memories and glossaries help carry approved terminology into later installments.

Copy link to this question

We design data types and labeling rules for the training objective, then process and review video, audio, images, and documents. Work includes parallel translation corpora, speech transcripts, image descriptions, video-and-question pairs, and specialist labeling. A small sample establishes the instructions and format before delivery of the agreed dataset volume.

Copy link to this question

Quality is reviewed at the design, processing, and final-dataset stages. We check label definitions and sample coverage, errors and omissions, alignment with the source, and the final volume and format. Error tolerances and re-review criteria are agreed at the start. Dataset quality and the performance of a model trained on it are evaluated separately.

Copy link to this question

It produces recurring content such as news summaries, market briefings, company updates, and educational videos. Approved scripts or data are turned into voice, subtitles, and graphics using templates with your logo, colors, and fonts. Specialists handle post-production and review. Multilingual versions and recurring delivery are scoped around video length, volume, and schedule.

Copy link to this question

No. Litmus is a review tool that flags passages and supplies source text for a human reviewer. Rules check length, whitespace, and repetition; contextual analysis considers blind-recruitment information, incorrect organization names, and offensive wording against the organization’s policy. Suspected plagiarism requires both similarity and consecutive matching. The reviewer makes the final eligibility and hiring decisions.

Copy link to this question

LETR K-Graph connects works, people, production companies, roles, and adaptation relationships across film, drama, music, webtoons, and novels. It retrieves connected facts as evidence, which an LLM organizes into a readable answer. Sources and review status can be inspected, and the system is designed to abstain when the required evidence is missing.

Copy link to this question

LETR WORKS helps you find content and scenes within your media and documents, then continue into translation, dubbing, and sharing. K-Graph focuses on facts and relationships among works, people, and adaptations, supplying evidence for AI answers. The right fit depends on whether you need to search within files or investigate connections among entities.

Copy link to this question

Start with the work you want to improve: LETR WORKS for content search and management, AI Signal for brand visibility in AI answers, or Preview Note for footage review. Translation and dubbing, training data, recurring video, application review, and content relationship research each have a corresponding solution. Share your current materials and desired output so we can recommend a suitable scope.

Discuss adoption and pricing →

Copy link to this question

Your team can operate a platform directly or use a service that includes specialist delivery and review. Integration with existing CMS or DAM systems and dedicated environments can be discussed according to the required functions and security needs. Options differ by solution; configuration is agreed around your materials, staffing, deliverables, and schedule.

Discuss adoption and pricing →

Copy link to this question

Pricing reflects the solution, functions used, processing volume, deliverables, and specialist review. Translation depends on languages, volume, and file formats; video production on length, quantity, and cadence; and dataset work on data type, processing complexity, and quality criteria. Video search indexing can be limited to selected assets. A quote is prepared from your sample and expected volume.

Discuss adoption and pricing →

Copy link to this question

We can assess the scope using your own work materials. Search uses sample content and queries; Preview Note uses footage and a document template; Litmus uses sample applications and review policies. Translation, dataset construction, and video production use small samples to align quality and delivery format. Sample scope, timing, and cost are agreed by solution.

Discuss adoption and pricing →

Copy link to this question

Repetition and review time matter as much as the amount of content. Recurring translation can reuse approved terms and translations, while recurring video can reuse brand templates. Compare processing cost, current working time, correction volume, and review effort on a sample before adoption. There is no single volume or installment count at which benefits apply to every project.

Copy link to this question

The solution pages distinguish available, partially available, and planned functions. LETR WORKS features such as short-form generation, a RAG chatbot, and external connectors are on the roadmap. The target-search API connecting K-Graph with AI Signal is also planned. Check the required features, integration scope, and delivery timing during scoping.

Copy link to this question

Customer content in LETR WORKS and the professional translation service is not used to train shared or general-purpose models. Approved translations and terminology are used in customer-specific translation memories and glossaries. Litmus applicant data is used for the relevant recruitment review, with retention and deletion set by the organization’s policy. Access, external processing, and dedicated-environment requirements are confirmed during scoping.

Copy link to this question

The Company page lists ISO/IEC 27001, 27701, 27017, and 27018 certificates covering AI-powered content localization SaaS. Listed versions of LETR WORKS and Gcon Studio received GS Grade 1 certification, and LETR API has a TTA quality and performance test record. ISO 9001 and 14001 are identified as certifications obtained in 2016. Check each certificate for its covered product or service and validity period.

Copy link to this question

LETR WORKS uses its own GPU inference for embeddings, search reranking, and OCR. Processing arrangements depend on the selected functions, so not all materials necessarily follow the same path. External processing scope, search-indexing targets, access permissions, and the operating environment are agreed during scoping.

Copy link to this question

We participated in NIA AI training-data projects for six consecutive years from 2020 to 2025, building 19 AI Hub datasets: 10 as lead organization and nine as a participant. Our record also includes five national R&D projects since 2017. The research achievements page lists project years, roles, and volumes. Data volumes should be read with their counting basis, including distinctions between source and labeled data.

Copy link to this question

The R&D page reports 10 registered patents in Korea and abroad as of December 2025. They cover AI translation and multimodal content processing, training data, and SaaS and workflow technology. The intellectual property page distinguishes registered patents, pending applications, and major trademarks.

Copy link to this question

It is Korean document retrieval R@10 measured in May 2026 on the MIRACL-ko evaluation corpus of approximately 300,000 documents. It describes recall within the top 10 search results, not overall service accuracy or a combined score for video and image search. The LETR WORKS page gives the evaluation conditions; adoption should also assess search quality using your own content and queries.

Copy link to this question

In the internal comparison dated September 1, 2026, K-Graph answered 109 of 119 answerable questions, with 108 correct. The 99.1% figure applies to answered questions; the correct-answer rate across all 119 was 90.8%. Questions about Korean content were built from K-Graph data, and comparison LLMs had web search disabled. These results do not guarantee performance across all domains or customer environments.

Copy link to this question

Use the contact page or email hello@twigfarm.net. Select the closest area, such as AI solutions, research collaboration, data and AX projects, media and content, or global partnerships. Explain the work you want to improve and the output you need so the relevant team can review your inquiry.

Contact us →

Copy link to this question

Helpful details include your current workflow, material types and expected volume, desired deliverables, languages, and timing. Include existing systems, security requirements, and a budget range if available. The form requires a name, email, inquiry type, and inquiry details. The relevant team responds within two to three business days after submission.

Contact us →

Copy link to this question

We can discuss national R&D projects, industry–academia research, technology validation and proofs of concept, and training-data collaboration. Review our research projects and technical areas, then choose Research Collaboration on the contact page and share your goals, required technology, participating organizations, and schedule.

Discuss research collaboration →

Copy link to this question