Hancom's PDF Tool Hits #1 on GitHub Trending · Issue #90

Hi, this is Oswarld. Did you land here after spotting my newsletter on Google Discover? If you subscribe, you’ll get fun, useful newsletters as fast as possible. 🙇

Opening

Hello, subscriber. On March 20th, a PDF data-extraction tool called “OpenDataLoader PDF v2.0” hit the #1 spot on GitHub’s overall trending chart. On the 21st alone, its star count jumped by 1,800, pushing the total past 7,000. At first, I had no idea who’d built it. Then I noticed a terms-of-service page and a Korean-language page and thought, is this a Korean developer? I was floored when I found out it was made by Hangul and Computer. Yes, that’s right — that Hancom.

I could just let this one-line piece of news slide by — “guess a Korean company built a decent open-source project.” But I see this as a meaningful inflection point for Korea’s IT industry.

Because Hancom has long been cited as the poster child of Korea’s IT “Galápagos” — the HWP file format, heavy dependence on government and public-sector contracts, negligible overseas revenue. Hancom Office has, for over 30 years, effectively sold only within Korea. While Microsoft Office held over 95% of the global office-suite market, Hancom’s entire game plan was defending roughly 30% market share domestically.

This is the first time Hancom’s technology has been directly validated by the global developer community — with no allowance made for being “a Korean company,” just raw benchmark numbers and star counts. Reaching #1 trending among roughly 400 million projects registered on GitHub is close to threading a needle.

Today, I’ll unpack how this was possible, and what this pattern means for other Korean companies — and for us as individuals.

📊 What the Benchmarks Say

Let’s start with the numbers. In OpenDataLoader’s own published benchmark, the average per-document scores for reading-order, table, and title extraction were as follows.

OpenDataLoader (hybrid): 0.907 ← #1
Docling: 0.882
Nutrient: 0.880
Marker: 0.861
Unstructured (hi_res): 0.841

I should be honest that this is a self-reported benchmark. That said, Hancom has published both the test data and the reproduction code on GitHub, so anyone who wants to verify the results can run it themselves.

What’s actually shocking here isn’t the score itself — it’s the speed. OpenDataLoader processes a page in 0.015 seconds. Marker, which shows similar accuracy, takes about 53.9 seconds per page. That’s roughly a 1,000x difference, and Marker requires a GPU, while OpenDataLoader runs on CPU alone.

Why does this matter? Say a company needs to process 1 million pages of PDFs — a task that would take Marker about 624 days, OpenDataLoader finishes in roughly 4 hours. For anyone building an AI training-data pipeline, this isn’t “a bit faster” — it’s a job that becomes possible for the first time. It opens the door to small businesses and solo developers who don’t have access to high-end GPU infrastructure.

It’s often said that 80–90% of a company’s operational data exists in unstructured formats like PDFs and documents, and feeding that data to AI has always been bottlenecked at the cleaning stage. PDF is a format optimized for rendering nicely on screen, not for extracting data. Where does a table start and end? In what order should a multi-column layout be read? How do you pull text embedded inside images? A tool that solves these problems accurately and fast is core infrastructure for the RAG¹ era.

The secret is its “hybrid engine.” Simple text is processed instantly via rule-based logic, and AI is only invoked for complex tables or multi-column layouts. That’s a different design from other tools that run a heavy AI model on every single page.

🏛️ What Hancom Really Built Over 35 Years

This raises a question: why Hancom, of all companies?

The open-source PDF parser market is already crowded with strong players — Docling (backed by IBM Research), Marker, Unstructured (a YC alum). And yet a 35-year-old Korean word-processor company outperformed tools built by global Big Tech and Silicon Valley startups on the benchmark.

This is where I think we need to reconsider what Hancom actually accumulated over 35 years. It’s Hancom’s own “document ontology²” — which text is a heading versus body copy, how to decompose a table’s rows and columns, how to reconstruct reading order in a multi-column layout. Hancom had 35 years’ worth of patterns for exactly these questions. Dong-hyun Park, Hancom’s head of technology application development, said in an interview that the company “built unrivaled technical capability on the back of experience processing document data for public institutions and numerous companies.”

What’s interesting is that Hancom signed an MOU with Duallab, a global PDF-technology specialist firm, in July 2025 and co-developed the tool with them. So it’s really the product of Hancom’s document-processing know-how combined with Duallab’s PDF expertise. It’s worth noting this isn’t purely Hancom’s own technology.

Still, the core point stands: Hancom’s own framework for “how to decompose a document” became the foundation for the global collaboration. The reason Duallab chose Hancom as a partner is precisely the weight of those 35 years.

Looking at the timeline, Hancom’s global moves have visibly accelerated over the past year: a strategic alliance with Tencent Cloud in December 2025, its first overseas contract for facial-recognition solutions with Japan’s CyberLinks in February 2026, participation in Japan IT Week in April, and now this GitHub #1. There’s a clear pattern here. Rather than shipping the finished Hancom Office product abroad, they’re peeling off the underlying technical assets field by field and releasing them through global partnerships and open source.

🌏 A New Formula for Escaping the Galápagos

It’s not that Hancom never tried to go global. Back in 2016, it announced plans to expand Hancom Office Neo into 5 regions — Latin America, China, India, the Middle East, and Russia. We all know how that went. Planting a flag with a finished office-suite product, in a market where Microsoft Office holds over 95% globally, was next to impossible.

But this time, with OpenDataLoader, the approach is completely different. Hancom didn’t take its finished product (Office) abroad. Instead, it extracted its core competency (document-decomposition know-how) and repackaged it in the exact form global developers need most — an open-source PDF parser. On top of that, they released it under Apache 2.0³, one of the most permissive licenses available, deliberately switching over from the previous MPL 2.0.

This is a genuinely important shift from a go-to-market perspective. It’s a move away from the “export the finished product” model, toward a model of “give away the core competency as standard infrastructure for free → monetize commercial services built on top of it.” It’s the same pattern as NVIDIA giving away CUDA for free to become the standard for GPU computing, or Stripe making its payment API developer-friendly to dominate payment infrastructure.

Hancom has, in fact, already laid out its next moves. It’s an “AI orchestrator” lineup running from data extraction (OpenDataLoader) → RAG connection (Hancompedia) → work assistance (Hancom Assistant). Once the free OpenDataLoader spreads globally, commercial add-ons and Hancompedia running on top of it are what generate the revenue. In Q2, MCP⁴** support and commercial add-ons are also slated to be added.**

That said, there’s no guarantee this succeeds. Topping GitHub trending and actually capturing global market share are different orders of problem. Docling has strong backing from IBM Research, and LlamaIndex has its own parser. Hancom clearly won round one, but the game has just begun.

Oz’s Lens

Having watched GTM strategies for a while now, the failure pattern for Korean B2B software companies going global has been almost identical: “translate the finished product that sold well in Korea into English and take it abroad.” Hancom Office followed this pattern, as did countless earlier attempts.

What makes Hancom’s move meaningful this time is that it broke that formula. Instead of taking its biggest, heaviest asset (the finished Office product) abroad, it extracted the core competency underlying that asset and converted it into a form global developers need right now.

The timing is exquisite, too. 2026 is the year AI agents genuinely started calling on external tools to get work done. When an agent reads a PDF, extracts a table, and generates an answer via RAG, whoever owns that first step — the PDF parser — determines the standard for enterprise workflows for years to come. The fact that Hancom deliberately switched its license from MPL to Apache 2.0 to lower the barrier to entry shows it was gunning for exactly that position. Getting in before the standard is set, and becoming the standard yourself — that’s textbook GTM.

I think this pattern carries big implications for Korea’s other Galápagos-style companies. Naver’s search know-how, Kakao’s messaging infrastructure, Toss’s financial UX, Coupang’s last-mile logistics. None of these can easily go global as finished services either. But if the core competencies underneath them get extracted and repackaged as standard infrastructure global developers can use, the story could change.

And the same applies to us as individuals. In the age of AI, our biggest weapon isn’t generalities — LLMs already know every generality on the internet. What’s valuable is the specific knowledge that exists only in our own heads: our own personal framework for organizing things. Like the precision Hancom built by focusing on documents alone for 35 years.

There’s one condition, though: you have to be able to convert that specific knowledge into a form others can use and put it out into the world. If it stays locked in your head, it isn’t an asset — it’s just experience.

✍️ Closing

Here’s today’s story in three lines.

Hancom OpenDataLoader’s #1 spot on GitHub is the first case of a Korean Galápagos company being directly validated by global developers — with no allowances made, purely on benchmark scores and star counts.
This was possible thanks to the document-processing “ontology” built up over 35 years, and this time, they extracted it as a core competency rather than a finished product and released it as open source. It’s a new formula from a GTM perspective.
This pattern applies to other Korean companies, and to individuals who’ve dug deep into a single field, alike. It’s an era in which anyone with their own framework for organizing knowledge has an overwhelming advantage.

So, subscriber — what have you spent the most time on over the past 10, 20 years? Whatever it is, there’s a good chance your own ontology is dormant inside it. The next question is how you pull it out into a form others can use — just like Hancom’s 35 years got cashed into the #1 spot on GitHub.

References & Further Reading

Primary sources

OpenDataLoader’s official site, benchmark page. Lets you directly compare accuracy scores across PDF parsers. The reproduction code and test data are also published on GitHub, so you can verify the results yourself.
Chan Park, “Hancom: ‘GitHub #1 Is the Fruit of 35 Years of Document Technology… It Will Become a Global Standard’”, AI Times, April 2026. An interview where Dong-hyun Park lays out the technical philosophy behind OpenDataLoader’s design in his own words — one of today’s key sources.

Background

“Best PDF Parsers for AI and RAG Workflows in 2026”, Firecrawl Blog. Compares the strengths and limits of competing tools like Docling, Marker, and Unstructured — useful for gauging where OpenDataLoader stands in the market.
Auer et al., “Docling Technical Report”, arXiv:2408.09869, 2024. The official write-up of the design philosophy behind Docling, Hancom’s toughest competitor — comparing it with OpenDataLoader reveals how differently the two tools approach the problem.
Hancom’s OpenDataLoader GitHub repository. The official repo you can download and try yourself — check the benchmark section in the README for comparison figures against other tools.

The author, Kwangseob Ahn, is a professor of business administration at Sejong University and lead consultant at OBF (Oswarld Boutique Consulting Firm). He teaches statistics and data analysis — business data management and business analytics — while leading GTM and AI strategy consulting in the field, designing the seam between technology and business. He has published academic research on a memory architecture for AI dialogue systems (HEMA) and runs Daily Arxiv, a daily curation of global AI papers. He holds a master’s from Korea University’s Graduate School of Technology Management and a KMBA. He is the author of Homo Brainless: The People Who Outsource Their Thinking.

Footnotes

RAG (Retrieval-Augmented Generation): a method where AI doesn’t just rely on what’s inside its own head (trained knowledge) to answer — it first retrieves relevant external documents and uses that content to inform its answer. Think of it like taking an open-book exam. That’s why how accurately you extract external documents (especially PDFs) is said to determine 90% of RAG answer quality. ↩
Ontology: a knowledge structure that organizes the concepts and relationships within a given field in its own particular way. Just as a library has its own system for classifying and organizing books, Hancom has spent 35 years developing its own way of organizing the domain of “documents” — what counts as a heading versus body text, how tables should be decomposed, and so on. ↩
Apache License 2.0: one of the more permissive open-source licenses. It allows nearly unrestricted commercial use, modification, and redistribution, and carries no obligation to re-publish modified code. Because it imposes the least legal burden on companies that adopt it, it’s one of the licenses most favored by global Big Tech firms and startups. ↩
MCP (Model Context Protocol): a standard communication protocol AI agents use to call external tools or data. Think of it like the USB port on a laptop — just as USB lets you plug in any device, MCP is the standard that gives AI a kind of “universal port.” ↩