Playment (YC W17) gives companies on-demand workers to analyze data using mobile | Hacker News Reader

Playment (YC W17) gives companies on-demand workers to analyze data using mobile

techcrunch.com

34 pointsrm290413 comments

Hi Guys,

I am Ajinkya (AJ) co-founder Playment. We started Playment as mTurk is highly unreliable in terms of data output quality or reliability. We focus on the standardisation of work across workers. In a nutshell we provide enterprise grade SLAs.

We deploy algorithms to generate high quality crowdsourcing output rather just grading the workers. Building a mobile workforce makes us faster than any other existing mTurk like solution. It's an fully managed platform where the task requester needn't worry about the task design or worker selection or deciding worker incentives. Everything is managed by the platform itself. Just share the requirements and data and we do it all.

The use cases that Playment target are cataloging for e-commerce, training data for AI (text classification, transcription, bounding boxes and image annotation).

We have written a whole piece on Playment vs. mTurk - https://playment.io/playment-vs-mechanical-turk

Would appreciate more comments and remarks. Hope you find this useful.

Looks promising. AI still has time to go before it can automate complex tasks. Also training models requires a lot of data, labelled training data is a huge bottleneck. Are there any other use cases you have planned for AI training data?

Sid, Cofounder @ Playment here.

Algorithms are only as good as the data that trains them. Besides having basic product features such as building bounding boxes on the images, identifying objects from images, context based text classification etc, we could also support multi step complex workflow. For example, for self driving cars, from a given image we could first find out if there are pedestrians present or not, then # of such pedestrians, then build boxes around it and finally tag them as males and females. To make it even more simple for our customers, we take over the complete ownership of building such complex workflows and ensuring quality.

Are you guys fixed on some specific usecases or trying to take care of almost all kinds of workflows?

Workflows have been built in such a way that it could handle multiple action on a single input. For example, we currently help a leading e-commerce based in India to do quality check o f the listings that are uploaded before they go live. The workflow involves more than 10 checks (image is blurry or not, background is white or not, description matches images, attribute check and so on) to be done on >3 images/product. All of this is handles by 1 workflow and output is given in few minutes.

Our platform can currently support services like image tagging, transcription, text tagging and bounding boxes. And workflows could be built using combination of these services.

At the end of the day, we have more than 100,000 skilled workers. We would love to hear more about various use cases that people would need definitely need humans for. :)

Amazing, this thing has the capability to grow almost like Uber. This should definitely beat Mechanical Turk once it reaches scale.

I was waiting for the "It's like mechanical turk, but for x" moment, but it didn't come. Perhaps a gamification twist given the name? Nope, apparently same same but more flexible for the workforce.

Akshay here, co-founder of Playment.

So there are quite a few differences, the primary being the quality & spam problem that mTurk has. We have a number of quality assurance models which help us promise enterprise grade SLAs to our customers.

Moreover, mTurk is limited to requesters from a few countries. We don't have any such restriction.

Also, Our product enables us to stitch together multiple tasks to model complex multi-step workflows so you don't need to take the hassle of setting up multiple tasks and compiling their results. Playment just does all of that for you.

We've put up a comprehensive piece on this: https://playment.io/playment-vs-mechanical-turk

This looks brilliant. How do you guarantee a fixed SLA ? I can see from your playstore app that your work force is non predictable as it is crowd-sourced ? Can you offer huge scale with limited downloads you have ?

With the 100k+ downloads we always have thousands of people active on the app around the clock. We've been able to tag more than a million images in a day; and that's a fraction of our total capacity. And we've acquired these 100k downloads with zero marketing spend, so there's enough scope to get more workers pretty easily :)

Looks quite interesting. This could perhaps give Turkers a much needed alternative to Amazon's Mechanical Turk.

Sounds like mechanical turk

Three major differences

1) SLA (Quality and turn-around times) assurance

2) Fully managed solution. Just send us the data and guidelines, we take care of the rest.

3) Support complex workflows

We wrote an elaborate piece on our differentiation against mechanical turk! Check it out on - https://playment.io/playment-vs-mechanical-turk

Reply on news.ycombinator.com