I accidentally built an AI platform (local ai server with local llm & workflows) for a friend's business that powers workflows and agents.
It's working so well that I have to explore this further. Today is the start of this journey. Since I love to follow founders that share their journey, I am trying this myself. @DominiqueCAPaul thanks for the inspiration!
I believe most companies are just scratching the surface of AI with simple chatbots and make/n8n workflows.
The real value is in deep, core automation - tackling the complex "AI Plumbing" to achieve massive efficiency gains
My long-term vision is to build the company that enables this for businesses, especially in Europe, to tackle our demographic and cost challenges.
But for now, I'm in listening mode. I want to understand how business leaders are thinking about this.
The goal for September 🍁
- Talk to business leaders about implementing AI
- Find a niche to bring value fast
- Build a network for what's next
Currently, I am mainly contacting my network of founders and business owners to get to meetings. This is fine for now, but I need to develop some sort of ideal customer profile in order to prospect more specifically. I hope that the next couple of meetings help me develop this.
Progress:
✅ 1 meeting done
🗓️ 1 meeting scheduled
"Hello World" for Agentic Workflows. 👋
Yesterday was a quiet breakthrough for Sophia, but a loud one for our architecture. We successfully ran our first end-to-end Agentic Action on the platform.
The Stack:
Local LLM
Local MCP (Model Context Protocol)
Local Supabase Database
The Use Case: We instructed the LLM to update a specific database entry (marking an email as done). Instead of hitting a pre-coded endpoint, the agent:
Understood the intent.
- Searched its available tools.
- Pulled the schema.
- Executed the update securely.
This creates a user experience similar to autonomous browser agents, but constrained within our application's specific security guardrails.
Why this matters: It allows us to move away from hard-coding hundreds of specific features and buttons. Instead of predicting every user need, we give them a controlled environment to execute tasks using natural language.
For me, this shifts the focus from "building features" to "building guardrails." It’s a completely different way of thinking about software design.
Happy to exchange ideas with anyone else building with MCP right now!
I am in BCN every year since 2010 as a tourist and never had any issues. Most crime is probably around pickpocketing in areas where you anyways would not go as a resident like Las Ramblas etc. Plus the surrounding of bcn (Girona, Costa Brava) is the best in the Mediterranean and super safe.