The 10th edition of the Sky Hackathon invites you to team up and take on the challenge of building intelligent conversational bots based on RAG

With the development and widespread application of large language models (LLMs), we are living in an era of technological innovation. To dig deeper and explore the potential of this technology, we invite you to attend NVIDIA's 10th annual Sky Hackathon. The core topic of this Sky Hackathon is "Building Intelligent Conversational Robots Based on RAG Technology Innovation". Participants are encouraged to design and build their own conversational bots using RAG (Retrieval-Augmented Generation) technology in conjunction with NVIDIA's SDK tools.

Join Sky Hackathon, 您将有机会

• Gain an in-depth understanding of RAG technology and master the core skills of building a conversational bot.
• Experience the power of the NVIDIA SDK for your own development efforts.
• Experience the power of NVIDIA NIM in generative AI and the process of deploying practices on edge devices.
• Learn from your peers and explore the possibilities of generative AI.
• Show off your creativity and prowess and win great prizes and participation certificates.
• 活动过程中，完成 NVIDIA DLI 线上课程《Building RAG Agents with LLMs》，还将获得 NVIDIA DLI 颁发的学习证书。
• During the event, writing study notes on CSDN or Zhihu will also get additional gifts.

Through your participation and efforts, Sky Hackathon will be a significant force in driving the development of generative AI technology. Looking forward to your joining and opening this feast of technology together!

关于 Sky Hackathon

Initiated and hosted by NVIDIA, the Sky Hackathon program helps university teams and AI developers deploy and optimize AI applications on NVIDIA Jetson edge high-performance computing products. Under the guidance of experienced GPU mentors, students will learn the deep learning-related application development and parallel computing skills required by the industry through hackathons, and stimulate the learning interest and innovation of the participating teams.

NVIDIA engineers will personally bring their understanding of the latest deep learning and edge computing, industry trends, and the latest technology applications and the latest development tools to the participating teams during the bootcamp.

The Sky Hackathon provides attendees with a unique opportunity to learn the hands-on skills needed for embedded deep learning development to accelerate and optimize their AI applications by using NVIDIA's latest programming models, libraries, and tools.

The entire event included a bootcamp and a hackathon, all online.

The top three finishers will receive a Jetson Orin NANO (4GB) development kit sponsored by tronorebo, the fourth through thirteenth place winners will receive a Jetson NANO development kit sponsored by tronorebo, and all team members will receive NVIDIA limited-edition badges and a gift.

*The finishing team will receive a certificate of participation

Event Sponsors

Hunan Chuanglebo Intelligent Technology Co., Ltd. (Chuanglebo) is an educational equipment company, related products mainly include: Raspberry Pi, NVIDIA Jetson series, etc., including thousands of products on science and education equipment.

【Official Website】:

[E-commerce platform]:

Event registration

A deposit of 100 yuan is required for this competition, and the 100 yuan deposit will be refunded after the event. After the team is successfully formed, the activity assistant will contact the team leader by email to collect 100 yuan.

WeChat identification QR code to register for this year's Sky Hackathon

A review of previous events

The first edition of the event can be accessed at:

https://mp.weixin.qq.com/s/NmKVftUs5diTrDRl2Rblmg

The second edition of the event can be accessed at:

https://mp.weixin.qq.com/s/1M50CStTGzQTY2ADuUpeYQ

Evaluation of the 2nd event instructor:

https://mp.weixin.qq.com/s/Sue6OV0eZyYlD2vvSLvwEg

The third edition of the event can be accessed at:

https://mp.weixin.qq.com/s/r-nd9zDbaYdrQ0nZzKl5vg

The fourth edition of the event can be accessed at:

https://mp.weixin.qq.com/s/df1QCJ5MRLd8CxrHkSylcg

The fifth edition of the event can be accessed at:

https://mp.weixin.qq.com/s/Rt9Xz_YmI9VsjL8tUsjuqg

The sixth edition of the event can be accessed at:

https://mp.weixin.qq.com/s/VzX3bjmXQ_el0W-Ilc8ltg

The seventh edition of the event can be accessed at:

https://mp.weixin.qq.com/s/nDNn6SszX14pE5KIqNMITg

The 8th edition of the event can be accessed at:

https://mp.weixin.qq.com/s/HGo0jZqHNyGW3UkJlMfaVQ

The ninth edition of the event can be accessed at:

https://mp.weixin.qq.com/s/Reu7-q3ujBg_zCwv54B3-w

Introduction to the tasks of this competition

The theme of this competition is "Building an Intelligent Conversational Robot Based on RAG Technology Innovation"

The Sky Hackathon competition aims to allow participating teams to gain an in-depth understanding and apply RAG (Retrieval-Augmented Generation) technology through practice, combined with large models, to build and optimize conversational robots independently. The specific tasks include the following parts:

◆ Learn the combination of RAG technology and large models:

Through an introduction provided by the organizer and a DLI (Deep Learning Institute) course (https://learn.nvidia.com/courses/course-detail?course_id=course-v1:DLI+S-FX-15+V1), participating teams will systematically learn how to use RAG technology to enhance the communication capabilities of conversational robots. Gain a deeper understanding of how RAG technology works and how it can be combined with large models to make conversational bots more intelligent.

◆ Learn to use the NIM platform to build and deploy conversational bots:

Teams will need to use what they have learned, combined with the NIM interface and SDK tools provided by NVIDIA, to build a conversational bot based on RAG technology on their own. After the construction is completed, the participating teams need to deploy the chatbot to the Jetson NX platform provided by the organizing committee. In this process, the participating teams need to ensure the stable operation of the robot on the platform.

◆ Optimized dialogue interface and added innovative features:

On the basis of ensuring the normal operation of the basic functions of the chatbot, the participating teams need to optimize the conversational interface to improve the user experience. Teams are encouraged to get creative and add custom innovations to the chatbot. These can be widgets that enhance the interactive experience, elements that add interest, or other designs that make conversational bots more useful and appealing.

◆ Project Presentation and Defense:

In the final round, teams are required to present their projects to the judges and the audience. This includes a demonstration of the chatbot's capabilities, an interface demonstration, and an introduction to innovative features. The participating teams also need to prepare for the defense session, and answer the judges' questions on the innovation, technical difficulties, and implementation process of the project, so as to show their in-depth understanding of the project and solid technical skills.

By completing the above tasks, the participating teams will not only be able to master the application of RAG technology and large models, but also exercise their hands-on skills and innovative thinking in practice, laying a solid foundation for future learning and career development.

Along the way, participating teams will experience the following:

◆ Build an LLM system with predictable interactions:

You'll learn how to put together a large language model (LLM) system that is able to interact predictably with users by utilizing internal and external inference components. This means that your system will be able to understand the user's intent and provide a coherent, accurate response based on context.

◆ Design Dialogue Management and Document Reasoning System:

You'll learn how to design a conversation management and document reasoning system that maintains conversation state and casts information into a structured format. This way, your conversational bot will be able to better understand the user's needs and maintain consistency throughout the conversation, providing the information the user needs.

◆ Efficient similarity query using embedding model:

You'll learn how to leverage embedding models for efficient content retrieval and conversation facilitation. With this approach, your conversational bot can quickly find the most relevant content for the user's question and provide valuable reference information in the conversation, thereby improving the user experience.

◆ Implementing, modularizing, and evaluating RAG agents:

Finally, you'll learn how to implement, modularize, and evaluate an agent based on Retrieval Enhanced Generation (RAG) technology that is able to answer questions about research papers in its dataset without any fine-tuning. This will make your conversational bot more accurate and efficient when dealing with knowledge in specialized areas.

◆ Experience the ease of NVIDIA NIM deployment:

Learn to use NVIDIA NIM (NVIDIA Inference Microservices), a cloud-native microservice optimized. Learn how NIM simplifies AI models into microservices and enables rapid deployment of models through containerization.

Overall, by participating in this competition, you will gain comprehensive skills in building intelligent conversational bots, from system design to implementation, evaluation, to optimization, to build a solid foundation for your career development or personal interests.

Schedule of events

Event format: online

Event Schedule:

Reasoning Platform (provided by the organizer)

硬件平台：NVIDIA Jetson Xavier NX（8G Memory）

操作系统：Ubuntu 20.04 L4T 64 位元桌面版

Development environment: JetPack 5.1.3

Teaching environment: Jupyter Lab 2.2

At the same time, the participating teams are provided with a real Xavier NX programming environment.

Grading Criteria

Practicality, industry value, innovation (40%)

Practicality (15%): Whether the solution actually solves a real problem or meets the needs of the real industry. Programs that are more practical will receive higher ratings.

Industry value (15%): Whether the solution has potential business value or can significantly improve work efficiency in the industry. Programs with high industry influence and wide application prospects will score higher.

Innovativeness (10%): The degree to which the solution is innovative in terms of technology, methodology or application. Unique and innovative ideas will be awarded more points.

Completeness of work (30%)

Basic function implementation (20%): Whether the basic functions required by the competition have been completed, such as the construction of chatbots and the integration of RAG technology. The complete implementation of each basic feature is the focus of the scoring.

Stability and reliability (5%): whether the work is stable during operation and whether it is prone to errors or crashes. Entries with high stability and reliability will receive higher scores.

Documentation and support (5%): Whether clear instructions, technical documentation, or supporting materials are provided. Well-documented and supported documentation will improve your work's integrity score.

Number of NIMs invoked and technical depth (30%)

NIM Interface Calls (10%): Scoring is based on the number of NIM interfaces invoked by participating teams. Titles that successfully integrate multiple NIM interfaces will receive higher scores.

Technology Convergence and Depth (20%): Evaluate how the participating teams reasonably integrate multiple large-scale model technologies into the solution, and evaluate the depth of technology implementation. Works with a high degree of technical convergence and in-depth implementation will score higher.

Extra points

User Interface Optimization/Beautification (+5%): Extra points may be awarded if the user's interface is more deeply optimized or beautified, providing a better user experience. The specific percentage of bonus points will be assessed according to the degree and effectiveness of optimization, up to a maximum of 5% of the total score.

Note: The proportions of the above scoring criteria can be fine-tuned according to the actual situation to ensure the fairness and reasonableness of the scoring. At the same time, judges should consider the overall performance of the work when grading, rather than focusing on only one aspect.

Event communication and Q&A

This Hackathon event will provide an exclusive WeChat group for event communication, and all team members who meet the entry conditions will be invited by the organizing committee to join the WeChat group.

NVIDIA will offer an exclusive event section on the Developer Community Forum:

About the captain of the participating team

Each participating team is required to designate a team leader.

The team leader of each participating team shall be responsible for the competition contact and prize collection on behalf of the team during the event. Team members are responsible for assigning their own rights and responsibilities within the team, and if there is any dispute (such as the method of prize collection and distribution), the organizer/executor will not be involved.

Q&A

Q: Can I participate?

The Sky Hackathon is currently aimed at college students and developers, but high school teams interested in participating are welcome as well. GPU programming experience is generally not required, but some Python and programming basics are preferred; He has a certain theoretical foundation for deep learning and neural networks.

Q: How do I register?

The Hackathon ethos is teamwork, and each team needs to choose a captain and give the team a name.

All team members should register separately and fill in the team name.

Each team is limited to 3-5 people. If there are more than 5 or 3 people in the team, we will consider the team invalid.

Q: Are there any rewards?

In general, Sky Hackathon is about cooperation rather than competition. The best prizes are a new level of performance in the code, time spent with experts, the ability to run on the Jetson platform, and an unforgettable experience that may bring you closer to a new paper or presentation. However, we do give out some symbolic prizes and certificates of participation.