-
Notifications
You must be signed in to change notification settings - Fork 1
MAG*NET Hyperlocal Context Maker and Engine
A central thesis of devices that are aware of their surroundings is that they should be able to gather context, infer meaning, make decisions, notify, etc. The current batch of consumer AI tech has black boxes for ingesting context and a black box for what comes out. To get control, this requires a variety of skills: data science, machine learning, AI ops, and maybe app development, plus a lot of testing. What I propose is we need a way to train context rapidly, without being a multidisciplinary domain expert. Turn on "Context Maker" while performing an activity. drag and drop the sources of context, turn dials on tuning knobs for speed/accuracy/power consumption/cost (if 3rd party systems are involved), select the AI where applicable. Then gather context. The system will tell you when it has enough data. The data never leaves your control. The system will also tell you how to verify, and will prompt you to verify accuracy. Publish your Context Model into the Engine. At this point, you need to wrap up the model in a service/app. This is where you would add triggers and actions (think of Zapier's workflow builder for zaps).
This document does not cover specification of implementation, just user stories.
- v1 Initial thoughts
- v2 Cleanup wording
- No accounts required for FANG companies to get started ... there are some use cases possible without spending money
- Leave it to the user to decide where to spend money or not
- You decide where your data goes. You can keep it local, or use your own cloud, or use 3rd party services (FANG included)
This gathers context from your smart watch/HR monitor. It will be able to know your baseline at rest, warm up, moderate exertion, full exertion level. Publish. Everyone's baseline will be different. An 80 year old man or woman will have a different level of fitness than a 20yo. So a generalized model may be a good starting point, but it will be customized for your use.
My service will use my phone+smart watch for gathering context. I use the model I created in step 1A. No data science and no PhD involved. I want it to post to Bluesky when I complete a workout and tag my running buddies. Very easy. Create the service on your phone, dial in parameters for accuracy, possibly add some windows of time to ignore.
There are a number of existing products that allow an average data scientist to train a model and deploy. This process is still somewhat geared towards the skills of data science / ML / embedded developer.
These tend to be geared towards B2B use cases, not B2C. We want to focus on something that is pure B2C use.