Cogniflow Questions

A: Hey,

Thanks for your interest.

One prediction is a call to any model, for example, 1 email or 1 tweet, you want to get the sentiment. In the case of images It is the same, 1 image - 1 prediction (an x-ray image for example). In the case of audio, 1 prediction corresponds to a max of 30 secs. If you want to process one audio of 2 minutes, it will require 4 predictions.

# Trainings per month are unlimited for the Appsumo deal. When you build a custom model it will take computation time there is a limit depending on your tier. As a reference, a simple text classification model of 1000-2000 records would take around 5-10 minutes to train.

We currently support only unstructured data, and the ability to export a model will be ready early next year. Our typical client does not want to deal with infrastructure/deployment activities, so we offer an API ready for every Cogniflow model, that’s a great advantage.

You can use it for client projects if you want if your plan limits are sufficient.

You can build your own AI models, which is our main feature, and you can connect through an API.

We are a distributed, talented and experienced team across the Americas.

Our platform differentiates from other AppSumo products because we offer an easy way to build your own AI models from text, image, or audio in the same platform providing a unifying experience and it gives our customers a lot of possibilities.

Thanks!

A: Hi jurkov,

You can only add users to collaborate on the object detection labeling process. This is the tutorial on how to add them:
https://docs.cogniflow.ai/en/article/how-to-label-images-for-an-object-detection-model-dcfg1y/

We are close to finishing a feature called "Organization," and when it is done (next few weeks), you will be able to add users to your account inside Cogniflow. We will notify all our users by email once it is ready.

Thanks,

A: Hi pandacat!

There are two possible ways to tackle this problem with Cogniflow. One is to train an image classification to detect a goal or not (at its simplest form) and then extract the frames of the video to check the model. The other option could be to train an audio classification to detect the game's narrator when he says "goal."

Please get in touch with us through our chat or email to explore more in detail if you are interested.

Thanks!

A: Hi mavener!

Yes, we are available for the 3 days extended to Plus Members.

If you want to use the same email account, you need to contact us to request the deletion of your free account before you can buy on AppSumo. Or you can use another email account for the free access.

Thanks,

A: Hi Vishnu,

We do not plan to come back in the short term.

Unfortunately, changes between Tiers are only allowed during our Appsumo campaign.

Thanks,

A: Hi Vishnu_VU,

At least for now, we do not plan to be included on black Friday. We will let you know in case our plans change.

Thanks,

A: Hi d4f,

If we understand your use case correctly, the quick answer is no with current features.
Depending on your object, one possibility could be to use our pre-trained Object detection model, which identifies around 80 types of everyday objects to extract that object automatically. If your object is specific to your domain, you can train a custom object detection model. A final step to know precisely if it is the same object is to use an Image similarity model, which we plan to include early next year.

If you can share some examples of images or audio, that would be great for us to give better advice. You can send those to info@cogniflow.ai.

Thanks,

A: Hi Iana9,

Yes, custom text models will work on the 50 languages list. Our pre-trained sentiment analysis works only for English, Dutch, German, French, Spanish, and Italian. But you can build your own sentiment analysis to support more languages.

Thanks,

A: Hi Taiff,

We will release a way to export models to be hosted in external servers just for the Tier 5 plan, but currently, it is not possible.

Our tech stack is on top of the python ecosystem.

Thanks,

A: Hi lf_fdz,

Yes, you can do that using our pre-trained model, "Extract text from image." Correct, one image is one prediction.

You can contact us at info@cogniflow.ai to get a quote for extra predictions.

Thanks,

A: Hi,

Currently, we do not have a way to build custom generative models. What you can do in your case is, for example, to identify if a new icebreaker question would be a good one or not. Or you can build a text classification model to determine which type of icebreaker a new input is.

Thanks,

A: Hi jurkoy,

The model is yours and only accessible to your account. We have backups in place every day. We will have a way to export the model soon for Tier 5 plans.

Please remember that one of the most important parts of every AI model is the data used to train the models, so if you keep your data safe, that would be one of the best backup strategies.

Thanks,

A: Hi there,

Unfortunately, you cannot build it for now. For your use case, you would require a tabular data-based solution (a set of columns with all your sensors' values).
Cogniflow works with unstructured data (image, text, audio). We plan to add capabilities for tab data in the future but it won’t be something to be included in the short term.

Thanks,

A: Hi Alejandro,

Yes, that is a great idea and we are considering to add this integration in the future. Do you want to use Cogniflow from Airtable similar to what you would do with Excel or Google Sheets, right?

Thanks,

A: Hi JN,

You can use Cogniflow to build some models that help you create a KYC solution. We will release a face similarity feature soon that will allow us to compare two photos of a person and validate their identity.
We do not support video as an input yet, but we will. Having said that, if you extract the frames of a video, you can use any image or object detection model to classify or identify things.

I know you already book a demo so let's discuss this in more detail there.

Thanks!

A: Hi,

Yes we plan to include in the near future a new model that supports many languages like Japanese. We do not have an exact date but we expect we can release it in the next few months.

Thanks,

A: Hi knot,

Yes, you can. You have two ways:

1) Use our Public Experiment, “Extract text from Image,” and get all the texts and from that, find using a regular expression or some other logic to identify the text you are looking for.
2) Train an object detection experiment to detect the text you want to extract. Then running that model, you will get from an image the coordinates of that specific text. You can crop that part of the image and then run the public experiment "extract text from image" to just get the specific text.

Can you please share some image examples with info@cogniflow.ai so we can give you better advice on how to tackle this problem?

Thanks,

A: Hi Lucian,

Currently, you can extract all the info from a receipt using our Public Experiment “Extract text from Image”:
https://app.cogniflow.ai/experiment/e3310034-1aae-ffd0-ab0c-aa9763eba215

You get a JSON response with a list of texts and their coordinates detected (from our test model page, you can download a CSV file with all the data extracted). If you want to build an end-to-end solution, you probably need some programming or other no-code tools to help you calculate the exact position of the text you want to extract, or maybe a regular expression can do the work. Different formats can make this a difficult problem. Some ways exist to train an AI to detect key value information that we are evaluating to add in the future to Cogniflow.

If you want to discuss this in a more personal way, please request a call here:
https://calendly.com/cogniflow/demo

Thanks,