Buildwise

Describe the robot you want to build, get a full component list with purchase links and compatibility analysis, see it rendered, and know exactly what will break before you spend a single dollar.

Comment

Inspiration

Every robotics team has lived this moment: the simulation works perfectly, the robot hits the real world, and it immediately fails in ways that feel obvious in hindsight. Friction was wrong. Latency wasn't modeled. Nobody tested what happens at 20% battery carrying max payload on a 3° slope.

We wanted to build the tool that prevents that moment — not after the robot breaks, but before a single part is purchased.

We also noticed that getting started in robotics has a huge barrier: beginners don't know what components to buy, can't tell if they're compatible with each other, and have no way to visualize what they're building until it's already built. We decided to solve all of it in one flow.

What it does

Buildwise is a full robot planning and risk intelligence tool powered by the Gemini API. It walks you through four steps:

  1. Describe your robot — tell Buildwise what you want to build and your budget in plain natural language
  2. Get your component list — Gemini generates a full parts list with real purchase links on Amazon, Mouser, and DigiKey, unit prices within your budget, and a compatibility analysis showing which components work with each other and which software frameworks they support
  3. See your robot rendered — Gemini generates an image of your fully assembled robot using the selected components, with a component legend identifying each part
  4. Run your Reality Check — Gemini predicts exactly what will break when your robot hits the real world: sim-to-real gaps, edge cases ranked by severity, and a prioritized fix list before you build

How we built it

We built Buildwise as a vanilla HTML + CSS + JavaScript application with zero build steps and zero backend, making it instantly runnable anywhere. All AI functionality runs through the Gemini 3.0 Flash model directly from the browser.

We used three distinct Gemini calls, each with carefully engineered prompts:

  • Component generation — a text prompt that forces structured JSON output with component metadata, compatibility tags, pricing, and purchase URLs
  • Robot render — an image generation call that uses the component list as context to generate a realistic 3D render of the assembled robot
  • Reality Check — a reasoning prompt that takes the robot description and component list together and returns a structured risk report with severity-ranked issues and actionable fixes

The interface follows an industrial mission-control aesthetic using IBM Plex Mono and IBM Plex Sans, designed to feel like a serious engineering tool rather than a consumer app.

Challenges we ran into

The biggest challenge was getting Gemini to return clean JSON every time — without markdown fences, without preamble, with the exact schema we needed — required multiple iterations and careful temperature tuning.

The image generation step was the trickiest to get right. Prompting Gemini to generate a coherent robot image that actually reflected the specific components selected required us to construct a very detailed visual prompt from the component list programmatically.

We also had to design the compatibility logic carefully — determining which components conflict with each other across hardware interfaces, voltage requirements, and software frameworks is genuinely complex, and we had to trust Gemini's reasoning while building validation around it.

What we learned

  • Gemini's multimodal capabilities are genuinely powerful when you combine text reasoning and image generation in the same pipeline
  • Structured JSON output from LLMs requires explicit schema enforcement in the prompt — the model needs to know exactly what you expect
  • A single-file no-backend architecture is underrated for hackathons — zero setup, zero deployment friction, demo anywhere instantly
  • The sim-to-real gap is a real and unsolved problem in robotics that even experienced teams struggle with — Gemini's reasoning about physical constraints is surprisingly deep

What's next for Buildwise

  • Real-time price checking via component distributor APIs
  • Export full report as PDF
  • Save and share sessions
  • Support for uploading existing schematics or simulation screenshots for analysis
  • Integration with ROS2 package suggestions based on selected components

Built With

Share this project:

Updates