One of my favorite hobbies was creating automation using sensors with a Raspberry Pi. I collected many versions of the Pi to create different projects, but, when LLM's came out I saw the light so to speak and in my mind saw the future. I decided to work on a system that would continuously monitor information and create an enterprise intelligence platform in a micro version on the pi because that would give me the proper constraints architecturally to deploy not only on the cloud but with edge as well. Privatized operational intelligence for sensitive data that can't be connected to the cloud. I decided 1. Always-on observation. 2. Local-first autonomy. 3. Signal detection instead of dashboards. 4. Prioritization 5. Memory - Append only ledger 6. Closed-loop measurement.
This allowed me to watch for change without asking and evolved into Mercury Engine AI: continuous observation, autonomous detection, signal normalization, attention scoring, persistent memory, and outcome measurement. Once I realized it's the same loop I knew it could apply to medical records, search data, leads, ads, business operations, security events, finance, creative and soo much more. So at that point I stopped thinking about wrappers and tools to solve single problems and started thinking about the observations themselves. If every meaningful change could be represented as the same kind of signal then one engine could watch any type of industry, prioritize what really matters, recommend and action, and measure whether the recommendation worked. That idea became the foundation of Mercury Engine AI
The first surface I created was an SEO vertical that connects to Google Search Console and monitors search data that gives recommendations how to improve search visibility for your business. During this time Mercury detected a search query that was getting impressions but no clicks, recommended rewriting the page title, and stored the baseline metrics autonomously so I made the change and forgot about it. Two weeks later Mercury Engine AI reconnected back to GSC on its own, compared the new data against the original baseline, and classified the outcome as a partial success after impressions increased by about 50%. It was the first complete loop where the system detected a problem, recommended and action, remembered what happened before the change and later measured the outcome without me having to remember to check.
As the architecture evolved I filed a provisional patent because to me the system itself was worth preserving and protecting. The platform is live and only me as architect and python coding the stack. I'm at the point now where it's built and system proven itself over and over again. There are multiple surfaces and proof of concepts. There are next to zero external users at this point and I'm here with the hope of feedback from the Tech community since I've been completely isolated from mainstream as I was building.
Does the idea of software that closes it's own Observation -> Recommendation -> Outcome -> loop feel meaningfully different from current dashboards and copilots or is it just novelty?
One of my favorite hobbies was creating automation using sensors with a Raspberry Pi. I collected many versions of the Pi to create different projects, but, when LLM's came out I saw the light so to speak and in my mind saw the future. I decided to work on a system that would continuously monitor information and create an enterprise intelligence platform in a micro version on the pi because that would give me the proper constraints architecturally to deploy not only on the cloud but with edge as well. Privatized operational intelligence for sensitive data that can't be connected to the cloud. I decided 1. Always-on observation. 2. Local-first autonomy. 3. Signal detection instead of dashboards. 4. Prioritization 5. Memory - Append only ledger 6. Closed-loop measurement.
This allowed me to watch for change without asking and evolved into Mercury Engine AI: continuous observation, autonomous detection, signal normalization, attention scoring, persistent memory, and outcome measurement. Once I realized it's the same loop I knew it could apply to medical records, search data, leads, ads, business operations, security events, finance, creative and soo much more. So at that point I stopped thinking about wrappers and tools to solve single problems and started thinking about the observations themselves. If every meaningful change could be represented as the same kind of signal then one engine could watch any type of industry, prioritize what really matters, recommend and action, and measure whether the recommendation worked. That idea became the foundation of Mercury Engine AI
The first surface I created was an SEO vertical that connects to Google Search Console and monitors search data that gives recommendations how to improve search visibility for your business. During this time Mercury detected a search query that was getting impressions but no clicks, recommended rewriting the page title, and stored the baseline metrics autonomously so I made the change and forgot about it. Two weeks later Mercury Engine AI reconnected back to GSC on its own, compared the new data against the original baseline, and classified the outcome as a partial success after impressions increased by about 50%. It was the first complete loop where the system detected a problem, recommended and action, remembered what happened before the change and later measured the outcome without me having to remember to check.
As the architecture evolved I filed a provisional patent because to me the system itself was worth preserving and protecting. The platform is live and only me as architect and python coding the stack. I'm at the point now where it's built and system proven itself over and over again. There are multiple surfaces and proof of concepts. There are next to zero external users at this point and I'm here with the hope of feedback from the Tech community since I've been completely isolated from mainstream as I was building.
Does the idea of software that closes it's own Observation -> Recommendation -> Outcome -> loop feel meaningfully different from current dashboards and copilots or is it just novelty?
mercuryengineai.com seolionpro.com/mercury/patent-outcome seolionpro.com/mercury/patent-core