# Ex-Disney APAC Founder Launches SYNAPUFF to Build Interaction Intelligence Base for AI-Native Terminals, Secures Seed Funding

- Category: Flash (快讯)
- Published: 2026-11-12T17:28:27.023Z
- Source date: 2026-11-12
- HTML: https://www.88z.ai/en/news/news-ex-disney-apac-founder-launches-syna-20261112
- Markdown: https://www.88z.ai/en/news/news-ex-disney-apac-founder-launches-syna-20261112.md
- Other language: https://www.88z.ai/news/news-ex-disney-apac-founder-launches-syna-20261112.md

SYNAPUFF, an interaction intelligence company for AI-native terminals, completed a seed round of several million yuan, invested by FIRST RULE VENTURES and Pudong Venture Capital. The team proposes the "P Framework" to enhance agents' contextual understanding and dynamic behavior adjustment, addressing limitations of traditional command-response models through non-verbal probing and feedback loops for more natural interactions.

## At a glance

- SYNAPUFF secured seed funding from FIRST RULE VENTURES and Pudong Venture Capital, focusing on interaction intelligence for AI-native terminals.
- Introduced P Framework, replacing traditional command-response with a probe-and-feedback loop to enhance agents' contextual judgment.
- Integrates existing visual and emotion models, using non-verbal signals for dynamic behavior adjustment and natural interaction.
- Optimizes evaluation of behavior timing and user responsiveness through multi-source data accumulation and bidirectional interaction feedback.

## Funding Background and Core Team

SYNAPUFF, an interaction intelligence project for AI-native terminals, recently announced the completion of a seed round totaling several million yuan, with investment from FIRST RULE VENTURES and Pudong Venture Capital. The company was co-founded by Li Shimiao and Dr. Fan Mingming. Li previously served as an early member at Disney's Asia-Pacific headquarters, working on classic IP derivative projects, later joined Shixiang Technology for hard-tech investment, and founded Yichu Technology, which focused on haptic feedback and multimodal interaction and received investment from Sequoia China. Dr. Fan is a tenured associate professor and doctoral supervisor at HKUST (Guangzhou), specializing in human-computer interaction and human-AI collaboration, and was listed in the global Top 2% scientists in October 2026. Combining expertise in entertainment experience, investment, and academic research, they aim to develop an interaction base that endows agents with higher "emotional intelligence."

## From Command-Response to Contextual Awareness

Traditional human-computer interaction relies on users actively issuing commands with devices providing passive feedback, a model showing limitations as AI becomes a native capability of terminals. While basic large models have enhanced task execution across various devices, accurately understanding user intent, judging interaction timing, and choosing appropriate responses remain industry pain points. SYNAPUFF positions itself as an interaction intelligence company, aiming not just to recognize emotions but to enable agents to understand contextual situations, judge boundaries, and dynamically adjust behaviors. For instance, a robot must combine environmental context, user status, and interaction history to decide whether offering water during a meeting is appropriate, thereby determining strategies for intervention, waiting, or withdrawal, shifting from unilateral execution to bidirectional perception.

## P Framework Technical Architecture

To address these challenges, SYNAPUFF proposes the "P Framework," transforming the traditional linear "perceive-understand-act" process into an interactive loop of "probe-receive response-decide-receive response-decide." This framework allows agents to initially express intent through lightweight non-verbal signals such as eye contact, body orientation, or voice, then decide subsequent actions based on user feedback. The architecture comprises three core components: an interaction dynamics model to predict user responses; a personality strategy to select actions based on roles, relationships, and behavioral constraints; and a non-verbal behavior generation model to dynamically schedule multimodal interventions. This design does not retrain all underlying capabilities but integrates existing visual and emotion recognition models with native non-textual action generation to form an independent interaction decision framework adaptable to various terminal forms.

## Data Accumulation and Evaluation System

Data is crucial for optimizing interaction intelligence. SYNAPUFF plans to accumulate interaction data by annotating controlled experimental data, utilizing public film and television materials, and deploying in real-world scenarios. Unlike traditional models that focus solely on task completion, the evaluation dimensions for interaction intelligence extend to the appropriateness of timing, user willingness to respond, and the natural continuity of interaction. This means model training requires feedback from bidirectional human-agent interactions rather than merely verifying correct command execution. Currently, the company is training relevant models, aiming to continuously optimize interaction strategies by comparing user feedback triggered by different actions, enabling agents to integrate more naturally into human life and work scenarios.

## Sources

- [36kr.com](https://www.36kr.com/%3C![CDATA[https://36kr.com/p/4066337448907656?f=rss%5D%5D%3E) (2026-11-12)

This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
