Mirra

Mirra is an InfoFi platform that combines content curation with model training, splitting users into creators and scouts who flag Web3 posts worth collecting on social platforms. Model training never lacks text volume but human-judged samples: scraped material mixes platitudes, machine output and advertising, and feeding it directly only replicates noise.

Screening runs in two passes. After a scout submits a post, account and text checks flag bots, templated writing and blocked keywords; passing posts are then scored for relevance, information density, actionability and originality, with creators and scouts each earning NLP points, capped at five submissions per person per day to limit farming. Points record marginal contribution to the model, and leaderboards rank accumulated contribution.