Specify a real life use case scenario to demonstrate TinyML as-a-Service

[doyu]

UoH incubation program, NEXUS, planned a Demo day, we'll present our TinyMLaaS there. We need some impactful demo scenario to get funding from investors.

The outcome of this item / task should be documented in Markdown, stored in the repo, and published on github page.

For example:

1. https://landing.ai/platform/?utm_campaign=mv_webpage_landinglens_product
2. https://docs.edgeimpulse.com/docs/edge-impulse-studio/learning-blocks/object-detection/fomo-object-detection-for-constrained-devices

In some big picture of this project / development, we've been working on "Hello world" so far. We will be switching from this "Hello world" to some more real life ML use case soon (e.g. Object detection) to demonstrate "TinyML as-a-Service".

Some of the TinyML as-a-Service advantages are: (1) TinyML can run with battery powered (2) TinyML models are changed on the fly

For example, you could monitor the number of cars in the parking area while in the evening you could count the number of people in the same area. If you have any good idea how to demonstrate our TinyMLaaS, please contribute to this discussion,

[robertmora]

Uploaded use-case-scenario.md

[doyu]

With today's discussion with Tuula, it might be better to nail down a little bit more how to demonstrate TinyMLaaS. Would it be possible for you, assignees, to polish this scenario to let people understand why TinyMLaaS is needed? IOW, I want to sell TinyMLaaS to investors ;)

[2laJ2]

First thing that comes to mind is, that this solution lets the user to choose and upgrade the (camera) equipment, so that there are more options. It should be pointed out that this is a "process model app", not just a new home surveillance camera software. This is just the first use case, with which the app is built.

Also, it could be pointed out that later on other sensors might be used, such as measuring temperature, air humidity, possible gas leak or perhaps carbon monoxide, which has no taste or smell, but is lethal. That could be valuable information for home owners.

Second, does this send the alarm directly to the smartphone without going through the cloud? Can the user access the live video footage (or camera memory card) directly without the cloud? Because if so, that would be an advantage. The commercial home surveillance camera systems charge monthly fees per camera for better services. How about TinyML? Will the customer have to pay monthly fees?

[doyu]

The commercial home surveillance camera systems charge monthly fees per camera for better services. How about TinyML? Will the customer have to pay monthly fees?

Our target customer is not yet endusers but they are some service operators. We plan to charge them by monthly subscripton to TinyMLaaS since we expect that they often need to orchestrate necessary services on-the-fly (e.g. per time of the day, or seasons, winter vs summer)

As @2laJ2 asked, if we aim at end users, we could charge per API call too. If a customer don't update the model, they won't pay any extra except that 1st installation.

As @anadis504 mentioned yesterday, this course has scheduled your demo day 3rd of March? It would be a good chance of alpha release of TinyMLaaS, where it would be demonstrated why TinyML is needed while there has been already Cloud / Edge ML solutions. We need a clear demo scenario for that. Then based on that clear scenario, we could prioritize the necessary features (BIs). For example, in demo, we could lower the backend features than frontend featuers since the purpose of this demo is not all about features but it should make people understand why TinyMLaaS is needed.

And here too ;)