Katusha: The AI Assistant That Transformed Email Management at MMC Innovation

When a technical consulting firm like **MMC Innovation s.r.l.** receives dozens of emails every day from clients, suppliers, public authorities, and partners, the biggest risk isn't losing a message — it's wasting precious time figuring out what to do with it. Before automation, every incoming email required: - careful reading - manual CRM lookup - creating folders on Register - entering data into the database - and assigning it to the right consultant This meant hours of administrative work every week, with a constant risk of human error. **Today, it's all automated.** I designed and implemented for **MMC Innovation** an intelligent system powered by **n8n** (running on a hardened VPS accessible only via Tailscale) and **Gemma 4**, a large language model running locally on their Mini PC with 128 GB of shared RAM/VRAM. The result is **Katusha**, the studio's internal operational AI assistant. ## How It Works 1. **An email arrives** in the company inbox 2. n8n intercepts it automatically 3. Katusha reads the subject, sender, and full content 4. Using a set of specialized tools, the AI decides in seconds what to do: - Is it spam or irrelevant? → Automatically moved to the "SPAM da controllare" folder - Is it a general communication from an existing client? → Archived in the client's dedicated folder - Does it involve a new project request? → Automatically creates the client (if new), the project, the contact person, generates the folders on Register, and moves the email to the right place - Does the project need to be assigned to a consultant? → It creates the assignment in the database and notifies the Lead or Support team 5. The team receives a clean, contextual notification by email — only when truly needed Everything happens **without anyone having to manually touch the inbox**. ## Technical Architecture - **Orchestration**: Self-hosted n8n with dozens of interconnected workflows (main flow + dedicated tools + sub-workflows) - **Intelligence**: Gemma 4 running locally via Ollama — zero API costs, zero risk of sensitive data leaving the company, and extremely fast responses - **Database**: PostgreSQL to manage clients, projects, contacts, and staff assignments - **Email Server**: Register.it with dynamic folder creation - **Security**: VPS accessible exclusively via Tailscale + additional hardening measures. No sensitive data is ever exposed The system is highly modular: each "Tool" is a separate workflow (create client, create project, move email, assign person, delete entity, etc.), and Katusha decides which one to call through tool-calling. This makes the entire flow extremely robust and easy to extend. ## Real-World Results - **Time saved**: Hours of administrative work reduced to minutes - **Zero email loss**: Every message ends up in the correct folder and in the database - **Automatic assignments**: Consultants are notified only when necessary - **Full privacy and control**: The AI runs entirely on the company's own hardware - **Scalability**: The same Mini PC already hosts multiple other AI models and automations without issues In practice, Katusha has turned one of the most tedious and repetitive tasks in a professional services firm into a smooth, traceable, and nearly invisible process for the team. ## Why It Worked So Well I didn't just build "a bot that replies to emails." I created a **complete operating system for the inbox** that speaks the same language as the studio: client and project codes, assignments, Register folders, and internal database logic. Most importantly, it was designed **together with the client**, with obsessive attention to MMC Innovation's real workflows. If you run a technical consulting firm, professional services company, or any organization tired of spending half the day "managing the inbox," this is the kind of automation that genuinely changes how you work.