My portfolio.
Proofreading.
The quick brown fox jumps over the lazy dog
Buffalo buffalo Buffalo buffalo buffalo buffalo Buffalo buffalo
James, while John had had "had", had had "had had"; "had had" had had a better effect on the teacher
A ship-shipping ship shipping shipping ships
That that exists exists in that that that that exists exists in
When you wait for the waiter, do you not become the waiter
Brief:
An ML development company asked me to proofread several articles on novel applications of their technologies. I checked the text for grammatical accuracy. Included is an excerpt of the original text, followed by my edit. Changes are shown.
Copy (original):
Challenge: Large retailers or specialized companies need to collect competitor prices to be able to offer best price for customers. In order to speed up and automatize the price collection process and improve the quality of the data it is proposed to take a photo of the grocery shelf with the necessary products. Then with the use of computer vision technology the system recognizes the products and identifies its name and price.
The difficulty of this process is related with:
1. A large variety of input data;
2. Complex customer business processes, e.g., the presence of very similar products from different manufacturers or adding/deleting new SKUs from monitoring scope;
3. Products that have similar appearance, but they might be different items;
4. Different packing volume within the same product item;
5. Different names of the same product items in different stores.
Solution: After a photo has been taken, the mobile app sends the image to the S3 storage. Then, using the image address on S3, the mobile app sends a recognition request to the server. The server creates a task for recognition in the database and queues the image address. There are "recognition workers" at the other side of the queue which download the image from the data storage and recognize it.
The results stored in a database. The mobile app retrieves the results from it. After that employee checks the recognized positions and in case of mistake enters the data manually. The corrected data can be used later to train the models additionally.
Recognition pipeline works as follows: The images are recognized by a detector, which finds areas of products and price tags. The found product areas are sent for product classification and the price tag areas are sent for price reading. Recognized products and price tags are further compared with each other using their spatial positioning in the image.
The found product-price pairs are further used in the packing volume classification. Here the hypothesis is used that a lower capacity item costs less than a higher one. At the end the recognized product and its price are returned to the client.
Copy (edited):
Challenge: Large retailers and specialized companies need to collect competitors’ prices to be able to offer the best price for their customers. In order to speed up and automate the price collection process and improve the quality of the data, one proposal is to take a photo of the grocery store shelf with the necessary items. Then, with the use of computer vision technology, the system recognizes the items and identifies their names and prices.
The difficulty of this process is related to:
1. A large variety of input data;
2. Complex customer business processes, e.g., the presence of very similar items from different manufacturers or adding/deleting new SKUs from the monitoring scope;
3. Items that have a similar appearance, but might be different;
4. Different packing volumes within the same item;
5. Different names of the same item in different stores.
Solution: After a photo has been taken, a mobile app sends the image to S3 storage. Then, using the image address on S3, the mobile app sends a recognition request to the server. The server then creates a task for identification in the database and queues the image address. There are "identification workers" at the other side of the queue which download the image from the data storage and identify it.
The results are stored in a database, from which the mobile app retrieves the results. After that, the employee checks the identified positions. In case of error, the employee can enter the data manually. The corrected data can be used later to further train the models.
The identification pipeline works as follows: Images are identified by a detector which finds the location and price tags of the items. The item locations are sent for product classification, and the price tags are sent for price reading. Identified items and price tags are further compared with each other using their spatial positioning in the image.
The found product-price pairs are further used in packing volume classification. Here, the theory is that a lower-capacity item costs less than a higher-capacity one. Finally, the identified product and its price are returned to the client.
An ML development company asked me to proofread several articles on novel applications of their technologies. I checked the text for grammatical accuracy. Included is an excerpt of the original text, followed by my edit. Changes are shown.
Copy (original):
Challenge: Large retailers or specialized companies need to collect competitor prices to be able to offer best price for customers. In order to speed up and automatize the price collection process and improve the quality of the data it is proposed to take a photo of the grocery shelf with the necessary products. Then with the use of computer vision technology the system recognizes the products and identifies its name and price.
The difficulty of this process is related with:
1. A large variety of input data;
2. Complex customer business processes, e.g., the presence of very similar products from different manufacturers or adding/deleting new SKUs from monitoring scope;
3. Products that have similar appearance, but they might be different items;
4. Different packing volume within the same product item;
5. Different names of the same product items in different stores.
Solution: After a photo has been taken, the mobile app sends the image to the S3 storage. Then, using the image address on S3, the mobile app sends a recognition request to the server. The server creates a task for recognition in the database and queues the image address. There are "recognition workers" at the other side of the queue which download the image from the data storage and recognize it.
The results stored in a database. The mobile app retrieves the results from it. After that employee checks the recognized positions and in case of mistake enters the data manually. The corrected data can be used later to train the models additionally.
Recognition pipeline works as follows: The images are recognized by a detector, which finds areas of products and price tags. The found product areas are sent for product classification and the price tag areas are sent for price reading. Recognized products and price tags are further compared with each other using their spatial positioning in the image.
The found product-price pairs are further used in the packing volume classification. Here the hypothesis is used that a lower capacity item costs less than a higher one. At the end the recognized product and its price are returned to the client.
Copy (edited):
Challenge: Large retailers and specialized companies need to collect competitors’ prices to be able to offer the best price for their customers. In order to speed up and automate the price collection process and improve the quality of the data, one proposal is to take a photo of the grocery store shelf with the necessary items. Then, with the use of computer vision technology, the system recognizes the items and identifies their names and prices.
The difficulty of this process is related to:
1. A large variety of input data;
2. Complex customer business processes, e.g., the presence of very similar items from different manufacturers or adding/deleting new SKUs from the monitoring scope;
3. Items that have a similar appearance, but might be different;
4. Different packing volumes within the same item;
5. Different names of the same item in different stores.
Solution: After a photo has been taken, a mobile app sends the image to S3 storage. Then, using the image address on S3, the mobile app sends a recognition request to the server. The server then creates a task for identification in the database and queues the image address. There are "identification workers" at the other side of the queue which download the image from the data storage and identify it.
The results are stored in a database, from which the mobile app retrieves the results. After that, the employee checks the identified positions. In case of error, the employee can enter the data manually. The corrected data can be used later to further train the models.
The identification pipeline works as follows: Images are identified by a detector which finds the location and price tags of the items. The item locations are sent for product classification, and the price tags are sent for price reading. Identified items and price tags are further compared with each other using their spatial positioning in the image.
The found product-price pairs are further used in packing volume classification. Here, the theory is that a lower-capacity item costs less than a higher-capacity one. Finally, the identified product and its price are returned to the client.
The quick brown fox jumps over the lazy dog
Buffalo buffalo Buffalo buffalo buffalo buffalo Buffalo buffalo
James, while John had had "had", had had "had had"; "had had" had had a better effect on the teacher
A ship-shipping ship shipping shipping ships
That that exists exists in that that that that exists exists in
When you wait for the waiter, do you not become the waiter
Brief:
A company specializing in GPU-powered audio production software asked me to proofread an upcoming LinkedIn post. I reviewed the post and checked for grammatical accuracy. Included is an excerpt of the original text, followed by my edit. Changes are shown.
Copy (original):
[redacted] made a real splash at the recent Audio Developer Conference 2022! It’s an important milestone for the startup, so I can’t be quiet about this fact and would like to thank my team, ADC organizers and the Conference participants for their sincere interest in [redacted] technology. My special thanks to XXX for making our participation possible.
Let me share some more details, facts, and emotions about how we managed to take part in this event. From the very beginning, starting from applying for the Conference and till its end, [redacted] technology became a subject of extreme interest from the professional community. We were even dubbed by colleagues as a ‘splash in the industry’.
To start with, we had to pass an anonymous poll by all the ADC members before applying for participation in the Speakers’ panel. We were selected to participate and moreover, all our presentations were highly appreciated by the audience.
Having performed with three presentations, we seemed to be the most active speakers this year, since we gave the highest number of presentations compared to the rest of the participants. Within the first working day of the Conference our workshop for the developers, demonstrating the platform’s full range of possibilities, was overcrowded.
Our offline spot was full of people interested in the [redacted] platform’s functions. The same situation was at our workshop in the online Jupiter Lab – designed to host 60 persons, it ultimately hosted a full capacity of 90 developers. As we were told later, our workshop was visited by over half of all the developers participating in the Conference.
By accident, our third presentation at the Conference coincided with Apple’s event held at the same time. So we didn't anticipate even half of the 400-seats auditory to be full but were pleased to see a fully-crowded room again.
This Conference gave us a lot of useful feedback which we are going to use further for product development. Our participation resulted also in ideas for new partnerships with some very progressive teams, we are eager to jointly implement them. I was happy to see many interesting people in person. Let’s meet next year at the same Conference again!
Copy (edited):
[redacted] made a real splash at the recent Audio Developer Conference 2022! It’s an important milestone for this tech startup, so I can’t keep this news to myself, and I would especially like to thank my team, the ADC organizers, and the conference participants for their sincere interest in [redacted]’s technology. My special thanks to XXX for making our participation possible.
Let me share some more details, facts, and emotions about how we came around to taking part in this event. From the very beginning, starting from applying for the conference and till its end, [redacted] technology was a subject of extreme interest from the professional community. We were even dubbed by colleagues as a “splash in the industry”.
To start with, we had to pass an anonymous poll by all ADC members before applying for participation in the speakers’ panel. We ended up being selected to participate and moreover, all our presentations were highly appreciated by the audience.
All in all, our speakers gave three presentations, and we seemed to be the most active speakers this year, since we gave the highest number of presentations compared to the rest of the participants.
Within the first working day of the conference our workshop for developers, where we demonstrated the platform’s full range of possibilities, was completely packed.
Our offline spot was full of people interested in the [redacted] platform’s functions. We saw the same situation at our workshop in the online Jupiter Lab – designed to host 60 people, it ultimately hosted a full capacity of 90 developers. As we were told later, our workshop was visited by over half of all the developers participating in the conference.
By happenstance, our third presentation at the Conference coincided with Apple’s event, held at the same time. Because of this, we didn't anticipate even half of the 400-seat auditorium to be filled, but were pleased to see a fully crowded room once again.
This conference gave us a lot of useful feedback which we are going to use further for product development. Our participation also resulted in ideas for new partnerships with some very progressive teams, and we are eager to jointly implement them. I was happy to see many interesting people in person. Let’s meet next year at the same conference again!
A company specializing in GPU-powered audio production software asked me to proofread an upcoming LinkedIn post. I reviewed the post and checked for grammatical accuracy. Included is an excerpt of the original text, followed by my edit. Changes are shown.
Copy (original):
[redacted] made a real splash at the recent Audio Developer Conference 2022! It’s an important milestone for the startup, so I can’t be quiet about this fact and would like to thank my team, ADC organizers and the Conference participants for their sincere interest in [redacted] technology. My special thanks to XXX for making our participation possible.
Let me share some more details, facts, and emotions about how we managed to take part in this event. From the very beginning, starting from applying for the Conference and till its end, [redacted] technology became a subject of extreme interest from the professional community. We were even dubbed by colleagues as a ‘splash in the industry’.
To start with, we had to pass an anonymous poll by all the ADC members before applying for participation in the Speakers’ panel. We were selected to participate and moreover, all our presentations were highly appreciated by the audience.
Having performed with three presentations, we seemed to be the most active speakers this year, since we gave the highest number of presentations compared to the rest of the participants. Within the first working day of the Conference our workshop for the developers, demonstrating the platform’s full range of possibilities, was overcrowded.
Our offline spot was full of people interested in the [redacted] platform’s functions. The same situation was at our workshop in the online Jupiter Lab – designed to host 60 persons, it ultimately hosted a full capacity of 90 developers. As we were told later, our workshop was visited by over half of all the developers participating in the Conference.
By accident, our third presentation at the Conference coincided with Apple’s event held at the same time. So we didn't anticipate even half of the 400-seats auditory to be full but were pleased to see a fully-crowded room again.
This Conference gave us a lot of useful feedback which we are going to use further for product development. Our participation resulted also in ideas for new partnerships with some very progressive teams, we are eager to jointly implement them. I was happy to see many interesting people in person. Let’s meet next year at the same Conference again!
Copy (edited):
[redacted] made a real splash at the recent Audio Developer Conference 2022! It’s an important milestone for this tech startup, so I can’t keep this news to myself, and I would especially like to thank my team, the ADC organizers, and the conference participants for their sincere interest in [redacted]’s technology. My special thanks to XXX for making our participation possible.
Let me share some more details, facts, and emotions about how we came around to taking part in this event. From the very beginning, starting from applying for the conference and till its end, [redacted] technology was a subject of extreme interest from the professional community. We were even dubbed by colleagues as a “splash in the industry”.
To start with, we had to pass an anonymous poll by all ADC members before applying for participation in the speakers’ panel. We ended up being selected to participate and moreover, all our presentations were highly appreciated by the audience.
All in all, our speakers gave three presentations, and we seemed to be the most active speakers this year, since we gave the highest number of presentations compared to the rest of the participants.
Within the first working day of the conference our workshop for developers, where we demonstrated the platform’s full range of possibilities, was completely packed.
Our offline spot was full of people interested in the [redacted] platform’s functions. We saw the same situation at our workshop in the online Jupiter Lab – designed to host 60 people, it ultimately hosted a full capacity of 90 developers. As we were told later, our workshop was visited by over half of all the developers participating in the conference.
By happenstance, our third presentation at the Conference coincided with Apple’s event, held at the same time. Because of this, we didn't anticipate even half of the 400-seat auditorium to be filled, but were pleased to see a fully crowded room once again.
This conference gave us a lot of useful feedback which we are going to use further for product development. Our participation also resulted in ideas for new partnerships with some very progressive teams, and we are eager to jointly implement them. I was happy to see many interesting people in person. Let’s meet next year at the same conference again!
The quick brown fox jumps over the lazy dog
Buffalo buffalo Buffalo buffalo buffalo buffalo Buffalo buffalo
James, while John had had "had", had had "had had"; "had had" had had a better effect on the teacher
A ship-shipping ship shipping shipping ships
That that exists exists in that that that that exists exists in
When you wait for the waiter, do you not become the waiter
Let me help you say what you want to say.
Fill in your contact info below for a free consultation. No job is too small.
contact@dls-english.com
English, made to order.
English, made to order.
All photos, images, logos, and fonts are either original works, used with permission of the company/individual, or free to use (unsplash.com, flaticon.com).