Pocket FM Launches ‘Sherpa’ AI Writing Engine After Reaching $500M ARR

Audio drama platform Pocket FM has unveiled Sherpa, a specialized fiction-writing AI trained on 5.5 billion hours of listener engagement data to power long-form serialized storytelling.

Audio drama platform Pocket FM has unveiled Sherpa, a specialized fiction-writing AI model trained on 5.5 billion hours of listener playtime data and minute-by-minute retention tracking. According to Pocket FM CEO Rohan Nayak, the implementation of Sherpa helped the platform double its annual run rate from $250 million to $500 million while increasing overall content production by 1,200% over the past year. To date, over 550,000 creators have used the tool to generate an annualized 2.6 million hours of content on the platform. Pocket FM reports that 10% of eligible writers earn over $200,000, three writers have passed $1 million in earnings within two years, and a single blockbuster series on the platform has generated over $100 million in revenue. Sherpa is designed to address specific limitations in standard large language models regarding long-form serialized fiction, particularly context window degradation and linear generation constraints. Its technical architecture relies on three primary components. First, a Narrative World Model utilizes an evolving semantic knowledge graph to track character states, plot dependencies, and core secrets to maintain continuity across hundreds of episodes. Second, a Hierarchical Story Planner decomposes narrative planning into discrete levels spanning seasons, arcs, sequences, episodes, and scenes, using dynamic replanning to update only affected future arcs when changes occur. Third, a Prose Engine trained on 1 billion tokens of Pocket FM stories uses specialized evaluator models to score drafts against a 40-item checklist, achieving an 80% to 90% benchmark alignment with human reviewers. The system features a 7-step creation pipeline that takes creators from initial concept interrogation and world building to sub-plot planning, interactive scene generation, automated editorial stress-testing, and one-tap voice conversion for audio production. Published works are directly distributed across Pocket FM's global platform, utilizing performance-based revenue sharing and automated localization to target international markets. Nayak stated that combining proprietary distribution with creation tools creates a compounding data loop, projecting that single-person studios using the platform will eventually build billion-dollar intellectual properties