Re:How Do Startups Founders Manage Social Media Content Crea...
by Jugurtha on Tuesday April 21, 2020 @12:45PM
(#22936646)
>But tell me something please, once you have a product and some users/clients, why don't you use more Social Media? To grow your brand loyalty and business growth? Social Media is pretty cheap.
We were mostly a tiny consulting company specialized in data projects, working only with large organizations that could afford us. We built custom solutions from end-to-end. In other words, we talk with an organization, listen to their problem, understand it, build acquisition lines to either get the existing data or capture it, sometimes manufacturing the hardware to acquire signals. Then build models (oooh, AI). Then build applications to use these models, either web or mobile applications, with user access control and admin features.
We handled everything and offered custom turn-key solutions to very large organizations and we did not need social media since there was more work than we could handle.
We did not need social media.
The good thing about this is that you have a tiny team that has worked on problems in banking, telcos, energy, health, and other domains and built a variety of systems. And once you have shipped for a client, you become their "goto" team for other projects. The bad thing is that it is taxing: not easy for an individual contributor to work on three projects simultaneously, and not easy to manage ten projects at the same time, have their code base in your head, contribute to them, know the contracts, and make the necessary decisions to move them forward either technically or otherwise. The bad thing is that you tend to go over similar steps (data, model, application) but the solutions were done with a "custom" solution mindset.
We took advantage of a change in the team's structure some time ago to start an internal project: a platform that could help us get data projects running and solve the problems we faced shipping these projects.
It handles data and prevents duplication, gives you a notebook with GPU support, tracks your models, params, metrics. Enables you to share your work with domain expert and enables them to play with it so we shorten the feedback loop (many ML practitioners complain they can't get their notebook/results in the hands of a domain expert easily). It also deploys your model so you could use it in the application seamlessly just by hitting an endpoint (i.e: a data scientist does not need to ask you to deploy it. Self service). And the applications are simply plugins: this is important because it allows to add "monitoring" as a plugin. It allows to add "Notebook" as a plugin. We're shipping features as plugins and it has simplified contribution.
Now: this is an internal platform and we don't have paying users yet. We have let around 30 students of our colleague to use it for their final year project, and we're too busy making sure they don't waste time. It is also good because they've hit some problems that we are fixing. They're benefitting in terms of compute power and data amount.
The platform is in pre-alpha. We are using it for our own projects to shrink the time to value for our customers. There are critical features that are lacking.
We will soon share a landing page for it listing these features and give access to certain people. We have already given access to some people to play with it. Some people working at FAANG companies have tried it, but they do not have the same problems everyone else is having (i.e: they have deployment engineers, data engineers, ml engineers).
We were mostly a tiny consulting company specialized in data projects, working only with large organizations that could afford us. We built custom solutions from end-to-end. In other words, we talk with an organization, listen to their problem, understand it, build acquisition lines to either get the existing data or capture it, sometimes manufacturing the hardware to acquire signals. Then build models (oooh, AI). Then build applications to use these models, either web or mobile applications, with user access control and admin features.
We handled everything and offered custom turn-key solutions to very large organizations and we did not need social media since there was more work than we could handle.
We did not need social media.
The good thing about this is that you have a tiny team that has worked on problems in banking, telcos, energy, health, and other domains and built a variety of systems. And once you have shipped for a client, you become their "goto" team for other projects. The bad thing is that it is taxing: not easy for an individual contributor to work on three projects simultaneously, and not easy to manage ten projects at the same time, have their code base in your head, contribute to them, know the contracts, and make the necessary decisions to move them forward either technically or otherwise. The bad thing is that you tend to go over similar steps (data, model, application) but the solutions were done with a "custom" solution mindset.
We took advantage of a change in the team's structure some time ago to start an internal project: a platform that could help us get data projects running and solve the problems we faced shipping these projects.
It handles data and prevents duplication, gives you a notebook with GPU support, tracks your models, params, metrics. Enables you to share your work with domain expert and enables them to play with it so we shorten the feedback loop (many ML practitioners complain they can't get their notebook/results in the hands of a domain expert easily). It also deploys your model so you could use it in the application seamlessly just by hitting an endpoint (i.e: a data scientist does not need to ask you to deploy it. Self service). And the applications are simply plugins: this is important because it allows to add "monitoring" as a plugin. It allows to add "Notebook" as a plugin. We're shipping features as plugins and it has simplified contribution.
Now: this is an internal platform and we don't have paying users yet. We have let around 30 students of our colleague to use it for their final year project, and we're too busy making sure they don't waste time. It is also good because they've hit some problems that we are fixing. They're benefitting in terms of compute power and data amount.
The platform is in pre-alpha. We are using it for our own projects to shrink the time to value for our customers. There are critical features that are lacking.
We will soon share a landing page for it listing these features and give access to certain people. We have already given access to some people to play with it. Some people working at FAANG companies have tried it, but they do not have the same problems everyone else is having (i.e: they have deployment engineers, data engineers, ml engineers).