LlamaIndex Launches neThing.xyz: Optimizing Text-to-CAD with RAG for 3D Generative AI

Summary:The LlamaIndex team has launched neThing.xyz, a platform that leverages Retrieval-Augmented Generation (RAG) to optimize the text-to-CAD workflow in 3D generative AI. By training a code generation AI on domain-specific languages for CAD, neThing.xyz converts natural language instructions into CAD model code, significantly improving the efficiency and accuracy of 3D modeling while reducing API costs by 80%. This innovation provides engineers with a smarter and more efficient tool for 3D design, s

---

A New Breakthrough in 3D Generative AI: The Birth of neThing.xyz

Over the past few years, generative AI has made significant strides in text (1D) and image (2D) domains, but 3D modeling remains a challenging area. Traditional 3D generative AI technologies excel in the digital realm but often fall short in practical applications such as engineering and manufacturing.

The Core Problem

The limitations of current 3D generative AI include:

• Insufficient Model Precision: Difficulty in generating models that meet the requirements of actual manufacturing.
• High API Costs: Complex 3D modeling tasks require a large number of API calls, resulting in high costs.
• Poor User Experience: Existing tools fail to meet the efficiency and precision needs of engineers.

The neThing.xyz Solution

To address these issues, the LlamaIndex team developed the neThing.xyz platform, featuring the following key innovations:

1. Code Generation AI Based on RAG:

   • By training a code generation AI on domain-specific languages for CAD, neThing.xyz can convert natural language instructions into CAD model code.
   • This approach not only improves the efficiency of 3D modeling but also reduces the number of API calls, significantly lowering costs.

2. Application of RAG Technology:

   • RAG technology combines retrieval and generation to provide AI with richer contextual information, enabling it to generate more accurate and relevant code.
   • In neThing.xyz, RAG technology reduces the average number of tokens per user query from 10,000 to 2,000, significantly enhancing system performance.

3. Real-World Applications:

   • neThing.xyz can handle various complex 3D modeling tasks, including curves, threads, pipes, and lattices.
   • The platform demonstrated its capabilities in the Too Tall Toby speed CAD competition, proving its potential in practical applications.

Technical Highlights

• Efficiency: Thanks to RAG technology optimization, neThing.xyz reduces API costs by 80%, saving users a significant amount of money.
• Intelligence: Utilizing code generation AI, neThing.xyz can automatically generate CAD model code, simplifying the workflow for engineers.
• Scalability: The application of RAG technology enables neThing.xyz to handle more complex queries and tasks, improving the overall performance of the system.

Future Outlook

The LlamaIndex team plans to further optimize neThing.xyz to make it faster, smarter, and cheaper. They will continue to use LlamaIndex to manage the RAG pipeline and look forward to collaborating with the open-source community to advance the development of 3D generative AI.

Recommendations for Developers

• Get Involved: Join the neThing.xyz community, share your feedback and suggestions, and help the platform improve.
• Explore Use Cases: Try applying neThing.xyz to different engineering and manufacturing fields to explore its potential value.
• Stay Updated: Keep an eye on the latest updates from LlamaIndex to stay informed about the latest developments in RAG technology and 3D generative AI.

Industry Impact

The launch of neThing.xyz marks a significant breakthrough in 3D generative AI for engineering, providing engineers with a more efficient and intelligent 3D modeling tool. This innovation not only enhances the practicality of AI in engineering applications but also points the way for future technological development.

---

Source: LlamaIndex Blog (2026-09-12)

— END —

Tags: #LlamaIndex #RAG #3D Generative AI #Text-to-CAD #neThing.xyz