Miguel Heredia

Key Skills

Software

Appen

Remotasks

Scale AI

Internal/Proprietary Tooling

Top Subject Matter

Chatbot machine transcription post edition (QnA of general topics).

Polygon labeling and categorization of CCTV recordings(People, cars, terrain).

Machine transcription post edition, audio wave segmentation and more. (Finances content).

Top Data Types

Audio

Image

Text

Top Task Types

Action Recognition

Polygon

Segmentation

Translation Localization

Freelancer Overview

I'm very good at audio related tasks, I have large experience making manual transcriptions and also MTPE. I have gained experience working on very exciting image annotation projects at Scale AI, where I mastered the platform's tool to draw very sharp polygons efficiently. From people to vehicles to terrain, I had a lot of experience annotating and labeling in that project. I have experience doing categorization tasks at Scale AI, where I used the platform to select the right labels for different products, for example.

Reviews (1)

Labeling Experience

Machine transcription post edition at AON Data

Internal Proprietary ToolingAudioTranslation Localization

I took a training course, then a test, and then I was given a green flag to start working. I would go to the work platform and start taking tasks from either the Gaming or Finances projects. The tasks were very draining as we had to correct the transcription, manually segment the audio waves, set speaker roles and label accordingly each segment where I had to filter music, game voices, personal information and any extremely offensive profanities. All that for 25 to 30 minutes long audios where each one had a lot of noise from the game and also very bad audio quality from the microphone of the speakers on the Gaming project. On the Finances project, the audios were shorter, from 10 to 16 minutes. The audio quality was cleaner, but these were podcasts where multiple people often overlapped their voices.

I took a training course, then a test, and then I was given a green flag to start working. I would go to the work platform and start taking tasks from either the Gaming or Finances projects. The tasks were very draining as we had to correct the transcription, manually segment the audio waves, set speaker roles and label accordingly each segment where I had to filter music, game voices, personal information and any extremely offensive profanities. All that for 25 to 30 minutes long audios where each one had a lot of noise from the game and also very bad audio quality from the microphone of the speakers on the Gaming project. On the Finances project, the audios were shorter, from 10 to 16 minutes. The audio quality was cleaner, but these were podcasts where multiple people often overlapped their voices.

2024 - 2024

Machine transcription post edition

AppenAudioTranslation Localization

The first few days I went through the guidelines and attended to two training meetings. After that, I started working on trial, I was given a milestone of 10 per day. The project had an Excel sheet full where the accuracy and total output of tasks completed was updated each day. Next day I had an accuracy of 90%, and so I was given 60 tasks, next day 120 and then, because I was consistently maintaining a high accuracy rate I eventually got to work on unlimited tasks per day, where I did 700+ tasks each day, each task being a minute long and me completing 100 tasks in about an hour. The tasks consisted on listening to the whole audio segment, correcting the machine transcription and labeling or segmenting the audio wave by the rules of the guidelines. Those tasks were then reviewed by QA where they approved or rejected just as a metric because the real QA was the number of my tasks that got approved by the client and that is what calculated the accuracy rate.

The first few days I went through the guidelines and attended to two training meetings. After that, I started working on trial, I was given a milestone of 10 per day. The project had an Excel sheet full where the accuracy and total output of tasks completed was updated each day. Next day I had an accuracy of 90%, and so I was given 60 tasks, next day 120 and then, because I was consistently maintaining a high accuracy rate I eventually got to work on unlimited tasks per day, where I did 700+ tasks each day, each task being a minute long and me completing 100 tasks in about an hour. The tasks consisted on listening to the whole audio segment, correcting the machine transcription and labeling or segmenting the audio wave by the rules of the guidelines. Those tasks were then reviewed by QA where they approved or rejected just as a metric because the real QA was the number of my tasks that got approved by the client and that is what calculated the accuracy rate.

2024 - 2024

Product categorization

Scale AITextEntity Ner Classification

After taking having completed the training course, I started working. The tasks consisted on visiting a website to study a product, then (on the work platform) I had to go over a list of different entities and different categories, browse and label the product accordingly. After that, I had to answer a few questions regarding the product.

After taking having completed the training course, I started working. The tasks consisted on visiting a website to study a product, then (on the work platform) I had to go over a list of different entities and different categories, browse and label the product accordingly. After that, I had to answer a few questions regarding the product.

2023 - 2023

Polygon image annotation

Scale AIImagePolygonPolyline

I did a training course, then I started working and depending on my performance I was given more tasks or not. I had only a certain amount of tasks I could do wrong during the first days, or I wouldn't be able to continue. Regarding the tasks, I was given images and I needed to outline entities from the pictures and label them. These varied from people to vehicles to terrain. For people, we had to outline and label the different parts of the body, which was extremely difficult in some cases. For vehicles, we had to identify the type of vehicle and annotate all the vehicles on the picture, except for those that were very small. Regarding the terrains, we had to go over a list of different surfaces, outline the surface only from everything that was on top of it and then label it accordingly.

I did a training course, then I started working and depending on my performance I was given more tasks or not. I had only a certain amount of tasks I could do wrong during the first days, or I wouldn't be able to continue. Regarding the tasks, I was given images and I needed to outline entities from the pictures and label them. These varied from people to vehicles to terrain. For people, we had to outline and label the different parts of the body, which was extremely difficult in some cases. For vehicles, we had to identify the type of vehicle and annotate all the vehicles on the picture, except for those that were very small. Regarding the terrains, we had to go over a list of different surfaces, outline the surface only from everything that was on top of it and then label it accordingly.

2022 - 2022

Education

Work History