▀████▀   ▀███▀
  ▀██     ▄█
   ██▄   ▄█    ▄▄█▀██▀███▄███▀████████▄  ▄█▀██▄  ▄██▀██  ▄██▀██▄▀████████▄▀██▀   ▀██▀
    ██▄  █▀   ▄█▀   ██ ██▀ ▀▀  ██    ██ ██   ██ ██▀  ██ ██▀   ▀██ ██   ▀██  ██   ▄█
    ▀██ █▀    ██▀▀▀▀▀▀ ██      ██    ██  ▄█████ ██      ██     ██ ██    ██   ██ ▄█
     ▄██▄     ██▄    ▄ ██      ██    ██ ██   ██ ██▄    ▄██▄   ▄██ ██   ▄██    ███
      ██       ▀█████▀████▄  ▄████  ████▄████▀██▄█████▀  ▀█████▀  ██████▀     ▄█
                                                                  ██        ▄█
                                                                ▄████▄    ██▀

Word clarity scoring for copywriters, marketers, and anyone who writes to be understood.

---

VernaCopy analyzes your copy word-by-word and scores it against psycholinguistic research — frequency, age-of-acquisition, concreteness, and familiarity data from real human processing studies. Not vibes. Not grammar rules. Actual data on how hard each word is for a human brain to process.

When a word scores low, VernaCopy suggests simpler alternatives that fit the sentence context — filtered by a local AI model and ranked by the same psycholinguistic data, so you're always swapping down in complexity, never sideways.

What's in it

Copy Scanner — Paste any block of copy. Every content word gets a clarity tier (S → D) and a score. Low-scoring words are clickable with inline replacement suggestions. Rescan after edits to track your score improving.

Word Lookup — Look up any single word and see its full breakdown: frequency score, age-of-acquisition, concreteness, familiarity, CEFR level, and alternatives ranked by clarity.

Word Compare — Put two words side by side and see which one is clearer and by how much.

How the scoring works

Each word is scored across four dimensions pulled from published psycholinguistic datasets:

• Frequency — how often the word appears per million words of spoken English (SUBTLEX-US)
• Age of Acquisition — the age at which most people learn the word
• Concreteness — how tangible vs. abstract the concept is
• Familiarity — how familiar the word feels to an average reader (MRC Psycholinguistic Database)

These are normalized and combined into a single clarity score. A word like use scores ~80. A word like utilize scores ~30.

How alternatives are generated

When a word scores C or D tier, the tool runs a three-step pipeline to find better options:

1. Synonym lookup — WordNet (OEWN 2025) for strict synonyms + SimplePPDB for corpus-derived simplifications with directional simplification scores
2. Semantic search — pgvector finds the closest words in the scored vocabulary by embedding distance, filtered to only words that score higher than the original
3. Context filter — a local LLM (qwen2.5:3b via Ollama) checks which candidates actually fit the sentence without changing the meaning, and can suggest additional simpler alternatives

The psycholinguistic data has the final say — no suggestion makes it through unless it scores higher than the word it's replacing.

Stack

Layer	Tech
Backend	Python 3.13, FastAPI, SQLAlchemy (async), Alembic
Database	PostgreSQL + pgvector, Redis
AI / NLP	Ollama (qwen2.5:3b), fastText embeddings, WordNet OEWN 2025, SimplePPDB
Frontend	React 19, TypeScript, Vite, Zustand, TanStack Query, SCSS Modules
Infra	Docker, nginx, JWT auth (access + refresh tokens)

License

AGPL 3.0