InfoDistill
An agentic research-summarization platform that extracts, classifies, and condenses technical articles using BART and zero-shot classification.
InfoDistill automates the whole path from raw source to readable digest: pull technical articles from multiple feeds, classify them by topic, and summarize them.
Classification is zero-shot rather than a trained classifier, which means new topic labels can be added without collecting labeled data or retraining, an important property when the taxonomy of "things worth reading" shifts every few months. Summarization runs on BART, chosen for abstractive quality on long-form technical prose rather than extractive sentence-picking.
The FastAPI backend composes the stages so extraction, classification, and summarization can each be run and debugged independently, which matters because these failure modes look identical from the outside: a bad digest could be a parse failure, a misclassification, or a bad summary.