Inspiration
"Right to Repair" movement is picking up steam across the world. Information about the repairability of a device should be easily available enabling consumers to make an informed choice on their next purchase.
What it does
We calculate the repairability score of a mobile phone on a scale of 1 to 10, based on the complexity of disassembling it.
How we built it
There are several YouTube videos that explain how to disassemble a mobile phone and replace/repair parts. However, very few summarize the complexity as a score.
We extract the transcript of a disassembly video and send it as a context to Gemini and ask a question to summarize the teardown into a repairability score from 1 to 10.
Challenges we ran into
Although Gemini does a great job of summarizing the main points of the tear down (even in different languages), the scores are not consistent. The base model is probably picking up language cues to determine the score, rather than actual steps.
This leads us to tune the base model with training data specific to mobile phone disassembly and their associated repairability score from reliable sources (like iFixit). We observed that the tuned model is more consistent across language variations, as demoed in the video.
Accomplishments that we're proud of
Helping consumers take into account of repairability before their purchase will have a big impact on sustainability. Consumer habits will also force manufacturers to make more repairable products which will help reduce e-waste.
What we learned
YouTube has a huge potential to be processed and made available to end users. AI technology like LLMs, Audio/Video models, etc. can accelerate tapping into this wealth of information.
What's next for Repairability Score Generator
Many high-quality videos (from smaller channels) do not have subtitles or transcripts. The next step would be to process video directly to summarize into steps and calculate the repairability score.
Log in or sign up for Devpost to join the conversation.