Keyword Tracker
Keyword Tracker ↗︎ keywords.kohjakob.com is a web app for AI-supported keywording of PDF documents. Every proposed keyword shows its confidence from a semantic and a syntactic model and best-matching passages highlighted in text. Custom free-text rules per keyword are checked by a local LLM, possible violations are flagged.
Made for the ↗︎ Public AI Hackathon 2.0 at ↗︎ AI:AT.
Tech Stack
- Python, Flask, React
- Semantic model: BAAI/bge-m3 document embeddings, logistic regression per keyword
- Syntactic model: TF-IDF, Ridge classifier per keyword
- Rule check: local Qwen3-8B on llama.cpp, yes/maybe/no score from token logprobs