Interactive 3D Object/Assembly Creation Website
Budget: $5,000 – $10,000 USD
The user should be able to create an account on the website. This account should hold the user’s past generations and files.
The user should interact with the website through a chat based conversation like ChatGpt. The user first inputs a prompt, like “Build a T-rex.” The website will create the T-rex and display it in the browser via three.js for an interactive 3d view. The user can then give feedback on the created object. For example, the user might say “Make the tail longer” in the chat. The website should then take this feedback and update the object accordingly. Another feedback might be “Make the arms moveable,” and the website will remember this. Then, once the user is satisfied, the website will use Onshape API to generate a file of an identical object that can be viewed in Onshape. The Onshape file will be created via Featurescript, which is then sent via the API to the part studio in Onshape. The reason I decided to use Featurescript is because of its versatility and ability to define custom parametric models that can easily be modified via parameter changes. This file should have the same functionality as the object in three.js, and it should have all the functions such as rotatable arms and the longer tail.
The website should be for assembly creation, mainly robotics. The LLM should be trained on MKCad specifically. The user can then specify which parts from the library to use. For example: “Create a simple drive chain that uses [motor name], [specific axle name], and 2 gears with 36 teeth.” The website should then create this with gears and axles that can spin via different joints like a revolute mate, etc.
The NLP should be a dynamic text to featurescript parser that understands what the user wants, both objectively and spatially. I want the NLP to be a trained open source LLM like Llama 4 to understand the relationship between text, assemblies, and featurescript. The LLM should be trained on a minimum of 10,000 Onshape MKCad assemblies for best understanding:
Assembly Complexity Distribution:
Simple (2-5 parts): ~2,000 examples
Medium (6-15 parts): ~5,000 examples
Complex (16+ parts): ~3,000 examples
Coverage of Joint Types:
Each joint type (bolt, pin, weld, etc.) needs 100+ variations
Common combinations need redundant coverage
Language Variation:
Each assembly needs multiple description variations (5-10)
Different terminology levels (technical, simple, domain-specific)
If object creation is too hard, we could create a diverse range of objects in a library. Then, if a request for one of the objects is made, we can start with a base model that is modified through a user's commands.
Finally, I want to add a section with my contacts like my LinkedIn, a small about me, etc.
The user should interact with the website through a chat based conversation like ChatGpt. The user first inputs a prompt, like “Build a T-rex.” The website will create the T-rex and display it in the browser via three.js for an interactive 3d view. The user can then give feedback on the created object. For example, the user might say “Make the tail longer” in the chat. The website should then take this feedback and update the object accordingly. Another feedback might be “Make the arms moveable,” and the website will remember this. Then, once the user is satisfied, the website will use Onshape API to generate a file of an identical object that can be viewed in Onshape. The Onshape file will be created via Featurescript, which is then sent via the API to the part studio in Onshape. The reason I decided to use Featurescript is because of its versatility and ability to define custom parametric models that can easily be modified via parameter changes. This file should have the same functionality as the object in three.js, and it should have all the functions such as rotatable arms and the longer tail.
The website should be for assembly creation, mainly robotics. The LLM should be trained on MKCad specifically. The user can then specify which parts from the library to use. For example: “Create a simple drive chain that uses [motor name], [specific axle name], and 2 gears with 36 teeth.” The website should then create this with gears and axles that can spin via different joints like a revolute mate, etc.
The NLP should be a dynamic text to featurescript parser that understands what the user wants, both objectively and spatially. I want the NLP to be a trained open source LLM like Llama 4 to understand the relationship between text, assemblies, and featurescript. The LLM should be trained on a minimum of 10,000 Onshape MKCad assemblies for best understanding:
Assembly Complexity Distribution:
Simple (2-5 parts): ~2,000 examples
Medium (6-15 parts): ~5,000 examples
Complex (16+ parts): ~3,000 examples
Coverage of Joint Types:
Each joint type (bolt, pin, weld, etc.) needs 100+ variations
Common combinations need redundant coverage
Language Variation:
Each assembly needs multiple description variations (5-10)
Different terminology levels (technical, simple, domain-specific)
If object creation is too hard, we could create a diverse range of objects in a library. Then, if a request for one of the objects is made, we can start with a base model that is modified through a user's commands.
Finally, I want to add a section with my contacts like my LinkedIn, a small about me, etc.