刘洋 (Liu Yang)

Optimized AI dialing strategies based on A/B testing with human agents, increasing connection rates from 50% to 65%.

Improved overall conversion rate from 5% to 8% by aligning AI product introductions with human sales strategies.

Analyzed human-AI collaboration data to identify root causes for lower conversion rates (3% vs. 8% for algorithm-only channels), leading to a 6% conversion rate increase in June by excluding historically saturated stores.

Proposed and launched the 'Cherry Blossom Season | Photo Shoot' special section within the mini-program's team-up feature, addressing high campus demand for photo shoots.

Acquired 49 new mini-program users within 3 days of launch through UI design and targeted operational promotion.

Increased daily active users by 36% compared to pre-launch, effectively activating user participation in the mini-program.

Proposed and launched an AI special effects camera, integrating visual, language, and image models to automate prompt creation for zero-threshold user experience.

Increased functional penetration by 50% compared to existing AI image modification features by enhancing community content diversity and reducing 'same-style' AI filter usage by 10%.

Collaborated with algorithm teams to engineer a C-side feature for one-click facial expression optimization, projecting a 5% increase in avatar photo download rate by independently building a ComfyUI workflow and pre-setting 16 common expressions.

Utilized VLM to reverse-engineer base image content into new prompts, increasing base image feature download rate by 1.5% and validating the technical feasibility for multi-modal generation scenarios.

Designed and launched the 'Theater' feed within the App's video page, introducing a new 'audio comic' content category by integrating internal AIGC production capabilities and novel resources.

Achieved a 2x increase in average playback duration for audio comics compared to other content categories by defining scenario and character preset prompts.

Developed and iterated a backend AI image matching tool for the 'Listen While Watching' experience, increasing user screen-on duration by approximately 16% for matched content compared to un-matched mode.