Project Info
[Modal] Inference Track (Inference Track Grand Prize: $5K in Modal Credits per person, paid visit (flight & hotel) to SF or NY Office with lunch with Modal team,. Inference Track Runner Up Prize: $1K in Modal credits, airpods for each team member.)
Mirage
This project did not submit a demo video on Devpost.
Motivation The future of AI-like robots and self-driving cars requires lots and lots of data. Current datasets are extensive, but they cover common senarios, like driving in a sunny or cold day. They overrepresent positive cases and underrepresent edge cases and failure cases, where things could actually go wrong. This leads us to believe that simulations of world models will be key to simulating edge cases, like driving when a tree hits the ground, to train more robust and safe models. Conviction is furthered by Waymo beginning similar research, seen here What did we do? Users will create prompts, such as "Generate a road that had a tree break and fall down on", a unique scenario that would be unlikely to be present in current datasets, but very valid and important to train a model on. An LLM hosted on Google Cloud will expand the user prompt Veo3 model hosted on Google Cloud will generate a video of the simulated road Modal hosting multiple h100 GPUs will run an open source world model / gaussian splatting algorithm to create a 3d representation of this world Future work (half completed at hackathon) Physics + simulations in 3D generated space Segmentation of objects in 3d representation space (enabling the moving of objects in the space, identification, and training)
Mirage
Won 1st place at TreeHacks Modal prize 2026!
Robots and self-driving cars need training data. SynSplatt synthesize 3D world-model training data covering hard-to-capture scenes - like driving in a hailstorm.
- https://github.com/kyan-yang/Mirage
- Try it at vercel
Created at TreeHacks 2026 by Shrey Birmiwal, Kyan Yang, Kevin Thomas, Adi Prasad
Motivation
- The future of AI-like robots and self-driving cars requires lots and lots of data.
- Current datasets are extensive, but they cover common senarios, like driving in a sunny or cold day.
- They overrepresent positive cases and underrepresent edge cases and failure cases, where things could actually go wrong.
- This leads us to believe that simulations of world models will be key to simulating edge cases, like driving when a tree hits the ground, to train more robust and safe models.
- Conviction is furthered by Waymo beginning similar research, seen here
What did we do?
- Users will create prompts, such as "Generate a road that had a tree break and fall down on", a unique scenario that would be unlikely to be present in current datasets, but very valid and important to train a model on.
- An LLM hosted on Google Cloud will expand the user prompt
- Veo3 model hosted on Google Cloud will generate a video of the simulated road
- Modal hosting multiple h100 GPUs will run an open source world model / gaussian splatting algorithm to create a 3d representation of this world
Future work (half completed at hackathon)
- Physics + simulations in 3D generated space
- Segmentation of objects in 3d representation space (enabling the moving of objects in the space, identification, and training)
Analysis
View
Metric
- 53
- 39
- 14
- 10
- 9
Figures cover GitHub contributors during the hackathon window. A co-authored commit counts in full for each author, so per-member totals add up to more than the whole-team figures.
Technology
- CSSIn code
- HTMLIn code
- JavaScriptIn code
- PythonIn code
- ReactIn code
- TypeScriptIn code
6 of 6 appear in the indexed code.
AI coding agents
- Claude CodeCommits
- CodexConfig
Detected from committed agent config files and commit authorship. Absence of a signal is not proof an agent was unused.
Codebase size
Source size
275 KB
Source files
37
Counts recognized source files only; vendored directories, binaries and lockfiles are excluded, so this is smaller than the repository on disk.
Repository
kyan-yang/Mirage
379 files · 515.4 MB · @ 9ddddb7
Structure
Interface
13 files · 3%Screens, components and styles rendered to the user.
Application logic
69 files · 18%Domain rules, services and shared utilities.
+2 more
Supporting
Layers are inferred from where files sit in the tree, not from reading the code. A project that names its directories unconventionally will read oddly here — open the file browser to check anything the diagram implies.
Languages
- Python61%
- TypeScript20%
- CSS10%
- Markdown5%
- JavaScript3%
- Shell0%
- Other (1)0%
Share of indexed source by file size. Binary and vendored files are excluded.
Dependencies
v3/frontend/package.json
npm · 11- @sparkjsdev/spark
- @types/three
- react
- react-dom
- three
- +6 more
v3/scripts/package.json
npm · 1- puppeteer
Declared in the repository’s manifests at the indexed commit. A declared package is not proof it is used, and runtime dependencies are listed first.
This project’s features have not been analysed yet.
Export this project's context (description, README, evidence, key source files) to chat with an AI agent elsewhere.