PROJECT SAMSEGI
ALGORITHMIC MODEL
AND SYSTEM ARCHITECTURE SPECIFICATION FOR HANGUL
SCRIPT PROCESSING
AND SYSTEM ARCHITECTURE SPECIFICATION FOR HANGUL
SCRIPT PROCESSING
I. Problem Statement:
Architectural Defect of Digital Representation
Architectural Defect of Digital Representation
Current Korean localization standards are inherently inefficient at the system level.
Historical Compromise: The KS C 5601 standard, adopted in 1987, hardcoded Hangul syllable combinations as monolithic "ideographs" (11,172 positions). This mechanical typewriter-era limitation was subsequently locked into Unicode and modern OS architectures.
Systemic Failure: Korean remains the only "tier-one" alphabetic system whose digital foundation relies on redundant dictionary lookups rather than combinatorics.
Operational and Financial Costs for Business
LLM Tokenization: An excessive "token tax" (a multiplier of up to 2.4x compared to English text). This drives up inference costs and forces reliance on
patented workaround algorithms.
Input Method Editors (IME): Persistent system-level composition bugs in modern CLI environments and editors (Claude Code, Warp, Ghostty, VS Code).
DBMS and Search: Data normalization conflicts break search indices, full-text search engines (SQLite FTS), and regular expressions.
Embedded Systems (IoT): The requirement to store heavy lookup tables for 11,172 syllables depletes the constrained memory of microcontrollers.
Conclusion: The root cause of the issue is not the linguistic complexity of the language, but a data structure defect that the industry is forced to mitigate with software workarounds.
Historical Compromise: The KS C 5601 standard, adopted in 1987, hardcoded Hangul syllable combinations as monolithic "ideographs" (11,172 positions). This mechanical typewriter-era limitation was subsequently locked into Unicode and modern OS architectures.
Systemic Failure: Korean remains the only "tier-one" alphabetic system whose digital foundation relies on redundant dictionary lookups rather than combinatorics.
Operational and Financial Costs for Business
LLM Tokenization: An excessive "token tax" (a multiplier of up to 2.4x compared to English text). This drives up inference costs and forces reliance on
patented workaround algorithms.
Input Method Editors (IME): Persistent system-level composition bugs in modern CLI environments and editors (Claude Code, Warp, Ghostty, VS Code).
DBMS and Search: Data normalization conflicts break search indices, full-text search engines (SQLite FTS), and regular expressions.
Embedded Systems (IoT): The requirement to store heavy lookup tables for 11,172 syllables depletes the constrained memory of microcontrollers.
Conclusion: The root cause of the issue is not the linguistic complexity of the language, but a data structure defect that the industry is forced to mitigate with software workarounds.
II. Inadequacy of Existing Approaches
Current solutions merely treat the symptoms while operating within a flawed encoding paradigm.
Unicode: the politically entrenched foundation of global IT. The original normalization
logic (NFC/NFD) cannot be changed, as any modification would break compatibility
with all existing data.
High-Level Frameworks: External libraries (such as libhangul or HarfBuzz) introduce redundant runtime computation layers, degrading overall performance without resolving the core issue at the kernel level.
Proprietary Binary Modules (SDKs): These lack transparency for security audits, complicate integration, and introduce critical supply chain attack risks.
Key Limitation: Any solution operating within the standard encoding paradigm is bound to constantly struggle with legacy data conflicts.
Target Solution Requirement: An architecture that sits above encoding conflicts requiring zero modifications to Unicode and causing zero destabilization to existing systems.
Unicode: the politically entrenched foundation of global IT. The original normalization
logic (NFC/NFD) cannot be changed, as any modification would break compatibility
with all existing data.
High-Level Frameworks: External libraries (such as libhangul or HarfBuzz) introduce redundant runtime computation layers, degrading overall performance without resolving the core issue at the kernel level.
Proprietary Binary Modules (SDKs): These lack transparency for security audits, complicate integration, and introduce critical supply chain attack risks.
Key Limitation: Any solution operating within the standard encoding paradigm is bound to constantly struggle with legacy data conflicts.
Target Solution Requirement: An architecture that sits above encoding conflicts requiring zero modifications to Unicode and causing zero destabilization to existing systems.
III.SAMSEGI Technology:
Combinatorial
Mathematical Core
Combinatorial
Mathematical Core
A return to the historical combinatorial foundation of the language (the Hunminjeongeum treatise, 1443) implemented at the machine instruction level.
Data Atomization: Complete rejection of pre-composed 11,172 syllables. Transition to 33 atomic numeric tags (one per basic grapheme). Deterministic Computation: Any syllable combination is calculated as a flat sum of tags in O(1) constant time.
Linear Addressing: Index-address calculation is performed without conditional operators (branchless execution), runtime tables, or expensive integer division operations.
Isolated Projection: A static table in rodata maps Unicode characters to SAMSEGI tags during initialization, completely eliminating lookups during core execution.
Result: Text stream processing is reduced to just 1–2 machine instructions. Featuring a zero-allocation design (no runtime memory allocation), zero overhead and absolute independence from external software.
Data Atomization: Complete rejection of pre-composed 11,172 syllables. Transition to 33 atomic numeric tags (one per basic grapheme). Deterministic Computation: Any syllable combination is calculated as a flat sum of tags in O(1) constant time.
Linear Addressing: Index-address calculation is performed without conditional operators (branchless execution), runtime tables, or expensive integer division operations.
Isolated Projection: A static table in rodata maps Unicode characters to SAMSEGI tags during initialization, completely eliminating lookups during core execution.
Result: Text stream processing is reduced to just 1–2 machine instructions. Featuring a zero-allocation design (no runtime memory allocation), zero overhead and absolute independence from external software.
IV. Seamless Integration
and Backward Compatibility
and Backward Compatibility
Eliminating conflicts with legacy systems at the memory design level.
Byte-Level Representation: The SAMSEGI tag layout in memory is physically
engineered as a valid string within standard encodings (UTF-8 / UTF-16).
Zero-Cost Translation: Data organization via union. At the hardware level,
the exact same memory region is simultaneously interpreted
as an integer tag for mathematical operations and as a byte array
for outputting to legacy interfaces.
Elimination of Conversion: Legacy files are read by the system without
prior modification; new output data is correctly recognized by legacy software.
Removal of the Normalization Layer: The mathematical model completely resolves
the runtime conflict between pre-composed syllables (NFC) and grapheme sequences (NFD).
Economic Impact on IT Infrastructure
Implementation requires zero database migration or data conversion.
Full backward compatibility with existing APIs and network protocols is maintained.
Costs for supporting and synchronizing dual text representations are heavily reduced.
Byte-Level Representation: The SAMSEGI tag layout in memory is physically
engineered as a valid string within standard encodings (UTF-8 / UTF-16).
Zero-Cost Translation: Data organization via union. At the hardware level,
the exact same memory region is simultaneously interpreted
as an integer tag for mathematical operations and as a byte array
for outputting to legacy interfaces.
Elimination of Conversion: Legacy files are read by the system without
prior modification; new output data is correctly recognized by legacy software.
Removal of the Normalization Layer: The mathematical model completely resolves
the runtime conflict between pre-composed syllables (NFC) and grapheme sequences (NFD).
Economic Impact on IT Infrastructure
Implementation requires zero database migration or data conversion.
Full backward compatibility with existing APIs and network protocols is maintained.
Costs for supporting and synchronizing dual text representations are heavily reduced.
OS | Windows | MacOS |
Program | Visual Studio Insiders 2026 | Xcode 2026 |
Operation Memory | 8 | 16 |
Build Format | Console App | Command Line Tool |
Stack | C++(98-present time) with union | |
Purpose of the application | decomposition of Korean syllables based on a mathematical model | |
Data for benchmark | kowiki-latest-pages-articles-multistream.xml(5.87 GB) | |
Syllables decomposed | 3.07 Billion/3.07 Billion found (100%) | |
Time(seconds) | 322 sec (9M per sec) | 118 sec (26M per sec) |
V. Delivery Model
and Risk Management
and Risk Management
Shifting risks from the technical domain to the legal one: an open architecture protected by a robust legal perimeter.
The technology is delivered as transparent source code under a strict Non-Disclosure Agreement (NDA). This eliminates the need for software obfuscation or binary protectors that degrade performance.
Enterprise-Grade Security for the Customer
Full Inspection Control: The customers InfoSec team can perform an independent audit of every line of source code to verify the absence
of backdoors or vulnerabilities (such as Buffer Overflow or Use-After-Free).
Trusted Build: Compilation occurs within the customers isolated corporate environment, completely mitigating supply chain attack risks.
Minimal Attack Surface: The minimal codebase footprint and lack of dynamic memory allocation guarantee runtime stability.
Intellectual Property Protection (Developer) The mathematical core
of the technology is secured by a legal contract backed
by strict financial liability mechanisms.
Any unauthorized use or duplication of code outside the scope of the contract is classified as a breach of corporate obligations, triggering direct legal and financial sanctions.
The technology is delivered as transparent source code under a strict Non-Disclosure Agreement (NDA). This eliminates the need for software obfuscation or binary protectors that degrade performance.
Enterprise-Grade Security for the Customer
Full Inspection Control: The customers InfoSec team can perform an independent audit of every line of source code to verify the absence
of backdoors or vulnerabilities (such as Buffer Overflow or Use-After-Free).
Trusted Build: Compilation occurs within the customers isolated corporate environment, completely mitigating supply chain attack risks.
Minimal Attack Surface: The minimal codebase footprint and lack of dynamic memory allocation guarantee runtime stability.
Intellectual Property Protection (Developer) The mathematical core
of the technology is secured by a legal contract backed
by strict financial liability mechanisms.
Any unauthorized use or duplication of code outside the scope of the contract is classified as a breach of corporate obligations, triggering direct legal and financial sanctions.
VI. Commercial Proposal
SAMSEGI is a verified low-level solution for optimizing
Korean language processing, surpassing existing standards.
O(1) execution at the hardware level (x86, ARM, RISC‑V).
No modification of legacy data is required; full system stability is preserved.
Native verified builds for Windows (Visual Studio) and macOS (Xcode). The code was developed based on the C++98 standard, taking into account modern development realities,
and is guaranteed to be compatible with C++11.
Functionality has been confirmed on Android up to the Android 2 emulator (benchmarking was not performed on the emulator).
Scales seamlessly from IoT microcontrollers to high‑throughput LLM tokenizers and search engines.
Korean language processing, surpassing existing standards.
O(1) execution at the hardware level (x86, ARM, RISC‑V).
No modification of legacy data is required; full system stability is preserved.
Native verified builds for Windows (Visual Studio) and macOS (Xcode). The code was developed based on the C++98 standard, taking into account modern development realities,
and is guaranteed to be compatible with C++11.
Functionality has been confirmed on Android up to the Android 2 emulator (benchmarking was not performed on the emulator).
Scales seamlessly from IoT microcontrollers to high‑throughput LLM tokenizers and search engines.
Terms of engagement (strictly under NDA):
1. Architecture audit. You receive the complete source code of the algorithms, the union, the mathematical core, and the stream processing logic, but without the tag table (SAMSEGI_TABLE). This allows an independent security audit and full verification of the processing logic transparency. The code is completely readable, yet non‑functional.
2. Performance demonstration. We run a pre‑compiled binary module on your data. You record speed, accuracy, and system load. The source code containing the tag table is not transferred.
3. Licensing and full source code delivery. Upon agreeing on commercial terms and signing a license agreement, you receive the complete source code, including the tag table, together with an unrestricted right of use in the target product. Code protection within your infrastructure becomes your responsibility.
1. Architecture audit. You receive the complete source code of the algorithms, the union, the mathematical core, and the stream processing logic, but without the tag table (SAMSEGI_TABLE). This allows an independent security audit and full verification of the processing logic transparency. The code is completely readable, yet non‑functional.
2. Performance demonstration. We run a pre‑compiled binary module on your data. You record speed, accuracy, and system load. The source code containing the tag table is not transferred.
3. Licensing and full source code delivery. Upon agreeing on commercial terms and signing a license agreement, you receive the complete source code, including the tag table, together with an unrestricted right of use in the target product. Code protection within your infrastructure becomes your responsibility.
Project completed
For inquiries and collaborations, please contact:
info@samsegi.com
*White Paper(En/Ko/Ru) + APPENDIX: EMPIRICAL PERFORMANCE VERIFICATION(En) is available
by prior agreement, exclusively for key corporate stakeholders, government officials, academics
and media representatives.
info@samsegi.com
*White Paper(En/Ko/Ru) + APPENDIX: EMPIRICAL PERFORMANCE VERIFICATION(En) is available
by prior agreement, exclusively for key corporate stakeholders, government officials, academics
and media representatives.
Gromov Arsenii
Russian software developer
creator of algorithmic word games for the Korean language
samsegiproject@gmail.com | samsegiproject@daum.net
creator of algorithmic word games for the Korean language
samsegiproject@gmail.com | samsegiproject@daum.net
The rights to the project are protected by copyright.