"Gumshuda AI" - GenAI powered Lost & Found App

Cool Project Ahead 🪭: How I made a Generative AI powered platform for lost and found belongings!
Its a very common problem-
You found a stranded item, want to return it, but don't know how.
This GenAI powered tool called "Gumshuda AI" Allows you to upload the image, add info on where you found it and provides a platform for the original owner to search for it.
What makes it cool?
*drum roll*
1 ) Zero Shot Algorithm: I never trained my ML model for this! The model which I am currently using recognizes objects, shapes, colors (around 400 Millions things, just out of the box)
2 ) Every search is semantic here. Whether its (text to text) search or (text to image) search, the model knows very well how to match your text search query with the image.
Hence uploading an image of a yellow apple and the text of "yellow apple" will point to the same thing.
3 ) You don't necessarily need an image of your lost item to be able to search it!
Upload the image if you have it,
or upload any reference image from the internet,
Even no image is fine! As already mentioned- image matching is based on semantic meaning of the search text.
How does it work?
Gumshuda AI uses clip-vit-large model to create a combination of sparse and dense vector embeddings and stores them in a vector database (in my case- pinecone). It utilizes the KNN (K-Nearest Neighbour) to find the closest matching vectors and returns the results accordingly.
A demo video is attached below.
I will soon be writing a detailed blog post about this tech stack- how can you add image search capablity to your RAG application and making this code open source too.
I have been building similar GenAI applications for more than a year now. Gumshuda AI was an ambitious project which could not get a proper closure. I pitched it to a couple of police stations as an internal tool but could not prove the value. In the end it remains a side project which also served as my final year's major project in college. Collaboration on this project are always welcomed.
See you in the next one! Regards
Cool Project Ahead 🪭: How I made a Generative AI powered platform for lost and found belongings!
Its a very common problem-
You found a stranded item, want to return it, but don't know how.
This GenAI powered tool called "Gumshuda AI" Allows you to upload the image, add info on where you found it and provides a platform for the original owner to search for it.
What makes it cool?
*drum roll*
1 ) Zero Shot Algorithm: I never trained my ML model for this! The model which I am currently using recognizes objects, shapes, colors (around 400 Millions things, just out of the box)
2 ) Every search is semantic here. Whether its (text to text) search or (text to image) search, the model knows very well how to match your text search query with the image.
Hence uploading an image of a yellow apple and the text of "yellow apple" will point to the same thing.
3 ) You don't necessarily need an image of your lost item to be able to search it!
Upload the image if you have it,
or upload any reference image from the internet,
Even no image is fine! As already mentioned- image matching is based on semantic meaning of the search text.
How does it work?
Gumshuda AI uses clip-vit-large model to create a combination of sparse and dense vector embeddings and stores them in a vector database (in my case- pinecone). It utilizes the KNN (K-Nearest Neighbour) to find the closest matching vectors and returns the results accordingly.
A demo video is attached below.
I will soon be writing a detailed blog post about this tech stack- how can you add image search capablity to your RAG application and making this code open source too.
I have been building similar GenAI applications for more than a year now. Gumshuda AI was an ambitious project which could not get a proper closure. I pitched it to a couple of police stations as an internal tool but could not prove the value. In the end it remains a side project which also served as my final year's major project in college. Collaboration on this project are always welcomed.
See you in the next one! Regards
My Linkedin: / whynesspower
My Github: github.com/why...