Technical Overview

Not search —
we link by function

BlackTurtle understands what a technology does, linking tech that does the same job even across classifications. This is an investor overview; core algorithms are protected.

Scroll down

Why it is hard

Great technology hides behind three walls

Cross-domain

A problem solved in turbine ceramics is needed in hydrogen storage — yet they never meet.

Cross-language

Half the world's patents are non-English. Korean patents are even more closed off.

Cross-corpus

Patents, papers, and government R&D don't talk to each other.

Engine

From one line of problem to an evidence-backed plan

Public
01

Ingest & normalize

We collect and normalize Korean patents and national R&D — the Korean corpus that starts the moat.

16,749 patents · 13,579 R&D projects
Core IP · Private
02

Function extraction (SAO)

We extract a "subject · action · object" function structure from each document — seeing tech by function, not class.

Extraction pipeline is private
Core IP · Private
03

Function vectorization

We embed body and "action/function" into separate semantic spaces so same-function tech sits close even when worded differently.

Embedding design is private
Core IP · Private
04

Fusion retrieval

We fuse body, function and keyword signals and lift "same function, other field" candidates by classification (IPC).

Fusion/re-rank weights are private
Public
05

Idea concretization

Personal ontology + retrieved evidence produce an executable plan. Every claim carries a patent/R&D source.

Same + cross-domain evidence + sources

Moat

The base, in numbers

0
Korean patents
0
National R&D
0
Technical docs
0d × 2
Function vectors

Data (a once-closed Korean tech corpus), function-extraction know-how, and a personal ontology that compounds per user — three layers of moat that thicken over time.

Defensibility

What we show, and what we protect

We show

  • What it does (function-based cross-domain matching)
  • Data scale and source coverage
  • Output shape and quality
  • The architectural big picture

We protect 🔒

  • Function (SAO) extraction & refinement
  • Embedding design & vector refinement
  • Fusion/re-rank weights & constants
  • Prompts & data-processing know-how

Competitors may mimic outputs, but the time and know-how of extracting and refining function on a Korean tech corpus is hard to replicate.

Concretize ideas on top of evidence.

Experience it