2025
Reedy — AI metadata for publishers
Chaptr.AI product: catalog metadata generation and optimization with editorial control.
I work with the Chaptr team on Reedy—an AI engine aimed at book catalogs: it supports keyword generation, THEMA subject codes, and SEO-oriented descriptions, together with RAG-style workflows, trend analysis, and multimodal capabilities (text-to-speech, speech recognition, image and video) where those fit publisher workflows.
Reedy is positioned around backlist metadata: refreshing older titles, tightening descriptions for search, and aligning subject data with common industry practice. The UI stresses human-in-the-loop control—you can apply changes per title or in bulk rather than handing the catalog fully to automation.
Reedy’s public pages address metadata managers, editors, and marketing managers as separate audiences on the same product.
Related publications — early generative NLP
Before today’s catalog-scale LLMs, a useful slice of my NLP work was sequence-to-sequence text generation for short Polish inputs: keyword extraction with a T5-style text-to-text transformer (arXiv:2209.14008) and a follow-up on transferable keyword extraction and generation at ICCS 2023 ( Springer chapter). Same era, different surface form: punctuation restoration for read Polish speech (PolEval 2021 Task 1, WikiPunct) and joint truecasing + punctuation for conversational ASR—early practice in cleaning model-facing text. See the PolEval punctuation project page for task write-up and paper links.