Marco

Inspiration

This project was inspired by the idea that managing databases and backend systems shouldn’t require deep technical knowledge. Many small businesses, warehouses, or medical offices still rely on manual record-keeping or outdated tools. We wanted to create something where a person could simply chat with an AI and have it:

Spin up a new project,

Manage the schema,

Insert/query data,

And handle everything behind the scenes automatically.

In short: turn natural language into a manager that maintain your work.

How We Built It

Frontend: Built with Next.js, Tailwind CSS, and shadcn/ui for a modern dashboard and chat experience.

Authentication: Supabase Auth to manage users securely.

Backend: Next.js API routes that forward user messages to Ollama running locally.

LLM Engine: Ollama with MCP (Model Context Protocol) tools for database operations.

Main Supabase: Stores global data — users, projects, chat history, and logs.

Project Supabase DBs: Each project has its own Supabase database for its unique schema and data.

Data Flow:

User sends a message in chat.

Message saved in Main Supabase.

Ollama interprets → calls MCP tools.

MCP tools connect to Supabase (Main or Project DB) and execute.

Response returned and logged.

What We Learned

How MCP can be used as a bridge between LLMs and structured databases.

The importance of keeping a single source of truth (Main Supabase) while isolating each project’s data in its own DB.

How to combine natural language interfaces with robust backend logic.

That users don’t care about schemas or queries — they just want results in plain language.

Challenges We Faced

Designing a system where multiple Supabase projects (one per user/project) could be dynamically managed by the AI.

Handling state between chat messages so that Ollama could correctly map a user’s intent to database operations.

Keeping the UX simple while managing fairly complex backend logic.

Integrating Ollama (local LLM) with Supabase through MCP, since this is still an emerging ecosystem.

Future Work

Support for multi-modal inputs (voice or image to project creation).

Auto-generating dashboards and analytics for each project.

Deployment of Ollama + MCP in a cloud-native setup for team collaboration.

Making the UI even more intuitive with drag-and-drop project management alongside chat.

Built With

• huggingface
• nextjs
• ollama
• openai
• react