I have been thiking around if this type of use cases for Thunk.AI is actually achievable or not.
Lets say we have a data base for the company products. Textual data (data base table) coupled with image files for each products. Let the AI learn and read those source data.
Then we take a photo for arbitrary product and pass image to the Thunk. AI. Then AI returns the best matched product details out of the source data.
Let s assume we are the sale persons facing the end customers and guide them in the show/model room. Just take a photo over there then capture the details for that product, without directing the IDs of that product.
If the company deal with thousands of the product being listed up, the seraching the catealog to find the details is troublesome. Just take a quck shot of the product, then AI tells us details of that product.
Super straightforward use cases, but I just wondered if this is possible or on the road map.
3 Likes
Hi Koichi,
I believe models for image similarity search like this have been around for a while.
About a decade ago (in the product that evolved into AppSheet), I was working on this sort of idea. But the challenge wasn’t the technology — rather, there wasn’t a compelling use case. In most commercial environments, products are tagged with barcodes or other things that enable easy lookup.
On the other hand, the new possibilities now (and enabled in Thunk) involve understanding an image and its contents, checking for similarity to or differences from a image, looking for instances of objects within the image, etc. So if you wanted to say take a photo of something and ask if it looks “normal” or “abnormal”, that would be something the models can do well and available via Thunk.AI for the “citizen user”
1 Like
Hi Praveen
Thank you for your positive feedback. If I understand correctly, it WILL be possible with Thunk.AI, but yet feasible for now?
If this type of functionality become available, could you guide us in terms of how to set it up? thank you.
Right now, you can add an image in your data (one of the columns types is Image). You would need to select an image file in your Google Drive for this, using the Drive file picker.
So for example, you can have a collection of Expense Receipts where each row contains an image column with a photo or scan of a restaurant bill.
Then you can ask your AI to extract information from it, or ask other questions of it (eg: “is there any alcohol being billed” or “what kind of cuisine is served at this restaurant”).
If it is an image of say a room, you can ask “How many chairs are in the room”. You can ask this interactively via a chat, or you can define an AI Formula column that automatically extracts and saves it.
What we havent’ enabled yet, but is easy to add, is to ask “how does this image differ from the expected standard image X”.
This possible use case is not really “extract text” type of use cases I assume.
Let me explain with some possible use cases. For instance, we assume the company is providing materials for home, such as wall paper, sanitary etc.
In terms of the wall page, we just take a photo randamly and pass to AI agent. AI agent will look into the companys product data globally and returns the product mathing best with what is appearing in the image.
Hope this help to understand what type of solutions I m looking for.
1 Like
Expanding this use case further.
Take a phote of wall paper. And ask AI agent, " could you list up the same type of the wall paper in terms of the color and mode?
Then AI look into the company product items globally and returns the answers.
2 Likes
Thanks for clarifying. Though that is a possible direction to grow, image database search is not at all our current focus.
We have built a system to host intelligent team projects. So you record work items in a thunk, assign them to team members, etc. And then the thunk helps get the work done, helps track and record the work, and verifies the information recorded.
So it is a spreadsheet + project system + GPT.
I think from your description you are looking for a SQL database + AI query language
1 Like
I think it is a logical request and one that will appeal to the retail industry. Especially for DIY stores. Let’s say you want to buy a faucet head, the faucet head you have may not be of the same brand, but the same model may have dozens of different brands. Take the picture, compare it with pictures with the same features in the database and list the products that match the appropriate features.
Maybe now is not the time, but it is a tool that should be added in the future.
It seems to me that over time, Thunk.AI will evolve into database tools (searching, checking, monitoring, advisors giving advice, etc.).
In other words, they will become trained, safe and all-knowing personnel for the company.
Of course, Thunk.AI will decide what should happen.
1 Like
With the exisiting solutions with Google, MS, Amazon or whatsoever AI solutions provider, it could be possible to achieve my use case by manually create the “MODELS” by paying bunch of money and time and resources.
My understanding was Thunk.AI is cut off such a long process and make the users to easily create the own models, regardless that it is based on Images, or text.
To appeal Thunk. AI as a new solutions to developper as well as citizen developer, you need to clarify more clearly the user can easily create the own model (normally, we need to educate/coach the own AI / argolism to create such agent.)
So far, this critical point is pretty much vague and unclear to me (ie. for the possible future users of Thunk.AI)
Then qustion is “What is the strong point with Thunk.AI? rather than building own AI model by manual?)”