Watch the TuringUnion promotional film and featured demos →
TuringUnion is a multi-agent autonomous education engine for personalized learning in the AI era. It follows the principle of “one learner, one university” and builds a personalized cultivation loop through learner profiling, path planning, multi-agent mentoring, dynamic classroom generation, and enterprise-level project training.
The system addresses four structural limitations in traditional education:
- Limited teaching formats
- Uneven teacher-student resources
- Rigid curriculum structures
- Weak connection between learning and practice
These limitations make it difficult for traditional education to achieve the following goals at the same time:
- Strong personalization
- High efficiency
- Scalability
TuringUnion is designed to move the learner from adapting to a fixed curriculum into a central position where goals, prior knowledge, learning rhythm, and practice performance dynamically organize the learning resources around them.
TuringUnion is built around three goals:
- Continuously understand learners through multi-turn dialogue and learning behavior collection.
- Dynamically organize cultivation through learner profiles, learning evidence, path generation, mentor scheduling, and classroom generation.
- Validate real capability through enterprise-level projects, code review, and structured evidence accumulation.
Every conversation, answer, exercise, project submission, and feedback signal enters the learning evidence system and informs subsequent path optimization, mentor routing, and capability evaluation. The core modules below describe the main capabilities in this cultivation pipeline.
TuringUnion follows a three-stage framework: intelligent cognition, intelligent cultivation, and intelligent verification. The system first models learner goals, capability estimates, feedback, and task constraints as a dynamic human node. It then organizes learning through personalized cultivation plans, multi-agent mentor teams, and interactive dynamic classrooms, and finally validates real capability through enterprise-grade project training and learning evidence modeling.
Learner profiling is the entry module of the system. The President Agent uses multi-turn dialogue to identify career goals, prior knowledge, capability gaps, and learning preferences. The profile evolves with learning behavior and feedback, forming a dynamic learner model.
The system generates staged learning paths from learner profiles. For example, an “Agent development engineer” path can be decomposed into multiple stages and nodes covering foundational knowledge, tool use, engineering practice, and project delivery. The path supports dynamic adjustment: it can compress when progress is fast and reinforce content when the learner is blocked at key nodes.
The mentor team consists of agents with different responsibilities. The system routes tasks according to the current learning objective and learner state, allowing different agents to handle explanation, practice guidance, Q&A, feedback, assessment, and support. This mechanism addresses the shortage and delayed response of mentor resources in traditional teaching.
The dynamic classroom generates teaching content from real-time learner interaction. Learners can interrupt, ask follow-up questions, or change focus during learning, and the system reorganizes the explanation around the current question.
This module is supported by TeachMaster, a self-developed teaching agent for course production workflows. TeachMaster covers syllabus planning, content organization, animation generation, voice explanation, and related processes. It has delivered 50,000 minutes of teaching, covering 42 first-level disciplines and 437 second-level disciplines. The cost per course is about RMB 1,600, roughly one percent of the cost of traditional recorded courses.
Enterprise-level project training connects learning with real industry needs. The system can integrate downstream enterprise project requirements. Learners complete development tasks in an online engineering environment, and the system automatically reviews code and project deliverables. The project results become trustworthy capability credentials and enterprise-facing project experience.
Watch the promotional film and featured demos on the video carousel site, or browse all files in the videos folder.
Participants: 22 undergraduate, master’s, and doctoral learners with some AI learning background but generally limited product analysis experience. Their learning goal was to grow into multi-agent product analysts.
Preliminary results show:
- Average score improvement of 25.14%
- Improvement across concept understanding, solution design, independent task completion, and enterprise project confidence
- User ratings above 4 points for path matching, AI feedback credibility, and job readiness confidence
The test provides initial validation for the “test—learn—test—practice” loop. The system can adjust learning paths based on testing and learning behavior, while accumulating traceable growth records.
TU0024 represents a beginner learner with limited AI and product experience but strong willingness to learn. TuringUnion starts with a clear path, structured courses, and enterprise-level practice to build foundational capability and deliver measurable progress.
TU013 represents an advanced learner with research foundations and prior AI project experience. TuringUnion fills knowledge gaps around the learner’s research goals and supports deeper practice through multi-agent collaboration and quantitative evaluation.
The two learners started from different baselines, yet both reached scores above 84 after learning. The beginner learner improved by 45.0 points, while the advanced learner improved by 24.5 points, indicating that the system adapts cultivation to different learner profiles.
TuringUnion targets personalized talent cultivation, AI education platforms, enterprise talent training, and job capability certification. The system connects course learning, mentor support, project practice, and capability validation into one learning pipeline, aiming to provide learners with more adaptive cultivation and make capability outcomes easier for industry stakeholders to understand and use.
- Website: https://www.turingu.cn
- Email: qianc62@gmail.com / aohan510@gmail.com