This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend
What I Built
I built SANKALP for my younger brother, who is currently around Class 6.
While helping him study, I noticed that he could read a chapter, look at a textbook diagram, and still struggle to understand the actual concept.
Take something as simple as a mitochondrion.
A textbook gives you a 2D diagram with labels like outer membrane, inner membrane, cristae, and matrix. I can explain what each part does, but the moment he has to imagine how those structures are actually arranged in three dimensions, it becomes much harder.
The same problem comes up with the human heart, engines, and many other concepts where understanding how different parts fit together matters more than simply memorising a picture.
At some point, instead of saying, "Look at the diagram again," we asked:
What if the diagram didn't have to stay 2D?
That became SANKALP.
SANKALP is an AI-powered 3D visualization system designed for students from Class 1 to Class 12.
A student enters a topic or query, and SANKALP attempts to turn that concept into something they can actually see, rotate, explore, and understand in 3D.
The goal isn't simply to generate a visually impressive model.
The goal is better understanding.
A student should be able to go from:"What is a mitochondrion?"
to:
"Now I can actually see how its different parts are arranged."
That is the problem SANKALP is trying to solve.
Live Demo
[https://sankalp2.vercel.app/]
Video Demo
[https://drive.google.com/file/d/14f5mbDT4jmmLcESnkNdyOZx6WiRcRr1Q/view?usp=drive_link]
The demo shows the journey from a student's query to a retrieved or generated 3D learning experience and interactive exploration.
GitHub Repository
Fumer057/SANKALP
The repository contains the complete full-stack implementation, including the frontend, FastAPI backend, retrieval pipeline, semantic validation, 3D asset handling, and generative fallback.
How I Built It
The core idea behind SANKALP is a hybrid retrieval + generation architecture.
We didn't want every query to trigger generation when a suitable resource already exists.
So the system first tries to retrieve relevant information and existing 3D assets.
Those candidate assets are then evaluated using an AI-based semantic validation layer to determine whether they actually match the student's query.
If a candidate crosses the configured confidence threshold, SANKALP serves it directly.
If no suitable candidate is found, the system falls back to AI-based 3D generation using Shap-E.
Retrieve when we can. Generate when we need to.
Query Understanding
The pipeline starts with a natural-language query from the student.
For example:
"Show me a 3D model of a human heart."
The system processes the query to identify the main entity, category information, and descriptive visual properties that can be used during retrieval.
Retrieval
SANKALP searches across multiple sources instead of depending on a single repository.
It can work with:
- Local SQLite-backed cached assets.
- Curated model sources.
- Sketchfab.
- Public CDNs.
- Three.js public repositories.
This retrieval-first approach allows the system to reuse an existing model when an appropriate one is already available.
Semantic Validation
Finding a 3D model is not enough.
A retrieved asset might technically match the search term while still being irrelevant to the actual request.
SANKALP therefore uses an AI-based validation layer to compare the candidate metadata with the original query and calculate a normalized confidence score.
Only models that satisfy the configured relevance threshold are selected for delivery.
Generative Fallback
When retrieval does not produce a suitable asset, SANKALP falls back to Hugging Face Shap-E to generate a 3D representation.
This means the system is not restricted to whatever models already exist in public repositories.
It has a second path when the retrieval path comes up empty.
Interactive 3D Experience
Once an appropriate asset has been selected or generated, it is rendered through an interactive browser-based 3D viewer.
The viewer is built using:
- Three.js
- React Three Fiber
- Drei
Students can rotate, inspect, and explore the model instead of trying to mentally reconstruct a three-dimensional object from a static textbook image.
Why Does Open Innovation Matter?
Open innovation is a big part of why SANKALP was possible to build.
The project combines several different areas:
AI
Natural-language understanding, semantic expansion, validation, and generative 3D.
Information Retrieval
Finding relevant knowledge and models from multiple sources.
3D Graphics
Rendering and interacting with models in the browser.
Backend Engineering
Connecting all of those pieces into one reliable pipeline.
Using open technologies allowed us to combine these pieces without building everything from scratch or depending entirely on one proprietary platform.
We can use FastAPI for the backend, React and Next.js for the application layer, Three.js and React Three Fiber for 3D rendering, SQLite for local caching, and Shap-E as a generative fallback.
That freedom is important for a student-built project.
We can experiment with the retrieval strategy.
We can change the validation threshold.
We can replace an asset source.
We can try a different model.
We can improve one stage without rebuilding everything around a single closed system.
Open innovation turns:
"Someone should build this."
into:
"Let's build it ourselves."
And for SANKALP, that freedom was essential.
My Agent Session
A large part of the development involved AI-assisted engineering.
The agent helped with architecture exploration, implementation, debugging, integration between the frontend and backend, and iterating on different stages of the AI and 3D pipeline.
The Hand-over
The reason I built SANKALP was not a benchmark.
It started with my younger brother struggling to understand a diagram.
I could explain the same concept five different ways.
I could show him another image.
I could draw it again.
But he still had to imagine the missing dimension.
That was the moment the idea clicked.
Maybe the problem wasn't that he needed another explanation.
Maybe he needed a different representation of the explanation.
With SANKALP, the question changes.
Instead of asking him to imagine what the inside of a heart looks like, I can give him a 3D representation and let him explore it.
Instead of asking him to memorise another mitochondria diagram, he can actually look around the structure
And that is the part that matters to me.
I didn't build SANKALP because I wanted students to see more 3D models.
I built it because I wanted them to understand more.
Future Vision
SANKALP is designed with Class 1 to Class 12 education in mind.
The long-term goal is to make difficult concepts across subjects easier to understand through interactive visualization.
Imagine a student asking:
"How does a combustion engine work?"
and exploring the engine in 3D.
Or:
"Show me how blood flows through the heart."
and following the flow interactively.
Or:"Explain the structure of a cell."
and seeing the entire structure spatially instead of staring at another flat diagram.
The bigger vision is to make interactive 3D visualization another layer of education, especially for concepts that are difficult to understand through text and static images alone.
My hope is that one day a student can simply type:
"I don't understand this."
And SANKALP can respond:
"Let's visualize it."
Prize Categories
- #devchallengesvg
- #weekendchallengesvg
- #hf26challenge
[Render]
Top comments (0)