Inspiration
- The project is inspired by a deep interest in the historical figure Su Shi, who was not only a literary master but also a gourmet and traveler, possessing a wealth of knowledge and life experiences.
- The goal is to use AI technology to enable people to converse with Su Shi across time, experiencing the joy of interacting with a historical figure.
- The aim is to provide a more vivid understanding of historical figures, beyond what can be learned from books and museums.
What it does
- The project utilizes the Eliza, NEAR AI and DeepSeek large language model (LLM) to create an AI agent that embodies Su Shi's appearance, personality, language style, and knowledge.
- Users can interact with "Su Shi" through text or voice, learning about his life, works, cuisine, and travel experiences.
- AIGC tools are used to generate realistic images and videos of Su Shi, enhancing the user experience.
- It provides an interactive platform, allowing users to communicate with an ancient celebrity.
How we built it
- The project is built on the Eliza agent framework, implementing agent prompts.
- NEAR AI model services are used to call the LLM interface, with DeepSeek v3 being selected.
- The frontend is developed using React, and the backend using Node.js.
- AIGC tools like 即梦 (Jimeng) AI and 可灵 (Kling) AI are used to generate Su Shi's images and videos.
Challenges we ran into
- Accurately replicating Su Shi's personality and language style.
- Generating realistic images and videos of Su Shi.
- Resolving compatibility and Chinese language support issues for voice broadcasting.
- Limiting Su Shi's knowledge to pre-Northern Song Dynasty era, which is different from typical LLM training.
Accomplishments that we're proud of
- Successfully creating an interactive "Su Shi" agent that can engage in conversations with users.
- Generating high-quality Su Shi images and videos, enhancing the user experience.
- Implementing basic voice dialogue functionality.
- Successfully combining historical figures with modern AI technology.
What we learned
- Gained in-depth knowledge of using the NEAR AI, DeepSeek LLM and the Eliza agent framework.
- Mastered the application of AIGC tools in generating character images.
- Learned how to address technical challenges in voice recognition and synthesis.
- Learned how to recreate historical figures using AI.
What's next for Su Shi Agent 苏轼智能体
- Further optimize Su Shi's language model for more natural and fluent conversations.
- Improve voice dialogue functionality, enhancing Chinese recognition and synthesis accuracy.
- Add more interactive features, such as "Su Shi" guiding users on "virtual tours" of historical sites.
- Integrate more large language models, to enrich the Su Shi agents knowledge.
- Hope to add more historical figures in the future.
Built With
- deepseek
- eliza
- jimengai
- klingai
- nearai
- react
Log in or sign up for Devpost to join the conversation.