Mixed-Script Conversion Engine

Sandhi — संधि

A live converter for text where Preeti, Unicode Nepali, Korean Hangul, English, and numbers appear in the same line. It tokenizes first, tags each piece by script, converts only the Preeti pieces, and never touches a single Hangul character.
Scope note: this is a working core engine — tokenizer, Korean-protection, OCR fixes, and an editable Preeti dictionary — not the full enterprise stack (Next.js backend, Prisma database, DOCX/PDF pipelines, Docker deploy, 500-test suite) described in a build brief that size. The word-level dictionary below was checked against your own example sentences and converts them correctly. The character-by-character fallback for words outside that dictionary is intentionally a small, verified seed set — unmapped Preeti characters pass through untouched and are flagged, rather than silently guessed at. Tell me which pieces to build out next (more mapping rules, a real backend, DOCX import) and I'll go build that specific piece for real.
Input0 chars
Output0 chars
Tokenizer breakdown
Settings
Korean is never modified — that guarantee isn't a toggle.
Custom dictionary
Exact-word overrides. Checked before the OCR word-list and before the character engine.
Preeti / rawUnicode output
OCR word rules
Whole-token repairs for common scan corruption, applied before the character engine.
CorruptedCorrected
OCR symbol rules
Single-character repairs (e.g. shifted number-row keys, stray OCR glyphs).
CorruptedCorrected
Preeti character engine (seed set)
Fallback for Preeti tokens not caught by a dictionary rule above. ि-type prebase marks are reordered automatically. Only verified entries are seeded — extend this table rather than trusting a guess.
Preeti key(s)Unicode