Inspiration
My kid watched Alexa turn a lamp on and off and asked a dangerous question:
“Can we make a car that Alexa or Google Home can control?”
So we started building.
Two hours later, we had:
- Consulted ChatGPT, Gemini, and Amazon
- Debated Arduino vs Raspberry Pi vs ESP (I still don’t really know what an ESP is)
- Assembled a bill of materials directly inside an Amazon cart that already contained winter clothes and a snow shovel
The project didn’t fail because it was hard.
It failed because we had to choose a battery.
Lithium batteries were expensive. One AI said AA batteries wouldn’t last. Another said we’d need eight of them for a few minutes of runtime.
Momentum died right there.
That’s how many projects perish — not from lack of ideas, but from decision analysis paralysis.
What it does
GemForge helps ideas escape decision hell.
It turns vague project ideas into clear, buildable execution paths:
- Defining opinionated bills of materials
- Mapping required tools and steps
- Reducing unnecessary choices so building can actually begin
GemForge Features
- Home Page: Initiate Project, Update Workbench, View and Open Projects
- Workbench: User can textually explain what they have or upload a picture of their tool bin - Gemini recognizes the parts / tools and updates the inventory which later gets used in Bill of Material authoring
- Create Project: User enters project title, desc and optionally paste images - Gemini builds a project context from those inputs, also builds initial version of DIY, PRO and KID bill of materials with detailed instructions and visuals
- Custom Bill: User can create custom bill of material
- Set Bill as Primary: User can select any of the bill of material as primary, and go to Execute that bill
- Execution: User can track parts statuses and manage the project progress as they work through the instructions
- Update Project: At any point user can update the project direction, which triggers Gemini to update all bill of materials and instructions
- Versions: Every change in project is saved as snapshot which user can review and restore from
GemForge focuses on momentum over perfection.
How we built it
GemForge was built on Google AI Studio Platform:
- It manages Projects, Workbench (Parts, Tools), Bill of Materials, Project Versions and status
- Uses Gemini 3 APIs for reasoning, image generation and image recognition to perform analysis
- Uses Supabase for persistence of projects, project versions and images
Challenges we ran into
- It is not apparent superficially but versioning is important feature here - ability to view and go back to earlier version of plan is crucial and improves user's confidence to move forward. Implementing versioning was challenging.
Accomplishments that we're proud of
- Turning a real-life failure into a product insight
- Building a system that favors execution over endless research
- Creating a real workflow that feels approachable, not intimidating
What we learned
Most projects don’t fail — they stall.
Even with powerful AI tools, building is still hard when:
- Information is fragmented
- Decisions are interdependent
- Every choice creates more uncertainty
Structure matters as much as intelligence.
What's next for GemForge
Next, we want to:
- Integrate with market places like Amazon, IKEA, Walmart so user / business can order parts right from GemForge
- Build / Integrate with knowledge base of wide range of parts, tools, costs and visuals
- Expand project templates, execution paths, smart version comparisons and better AI assistance in execution
Conclusion
GemForge is an attempt to help ideas escape decision hell and make it to execution.
Because building should be hard for the right reasons—not because we couldn’t choose a battery.
Built With
- gemini-api
- gemini-image-generation
- react
- supabase
- typescript
Log in or sign up for Devpost to join the conversation.