← Back to projects

Ramona

An AI personal assistant project exploring memory, automation, integrations and practical ways to make recurring AI workflows more efficient.

Ramona: integrations feed an assistant core; a SQLite store holds memory and state and is read before every model call.
‹RRamona ✨bot
Message…
Scroll inside the phone ↕  · 
Three screens of Ramona on Telegram: the introduction, the filtered inbox and the day at a glance.
Ramona as it is actually used, on Telegram. A recorded conversation: the content shown is illustrative, nothing runs live.

Why I built it & impact

Built to stop repetitive, low-value AI calls from eating into daily focus — email triage, calendar checks and job-search tracking were each their own manual chore. Ramona consolidates them into one assistant, cutting the number of daily manual check-ins and turning scattered tasks into a single conversational flow.

Built with

Anthropic Claude · n8n · Gmail API · Google Calendar · Tavily · SQLite

The challenge

Build an assistant that can handle recurring tasks without turning every interaction into a heavy, repetitive model call.

What I explored

  • Python and Claude API integration
  • SQLite-based memory and state
  • APScheduler for recurring tasks
  • Batching, token minimisation and prompt caching

What I built

A modular assistant architecture designed around memory, scheduled workflows and controlled model usage. The project became a practical sandbox for thinking about AI systems as products: not only prompts, but state, reliability, cost and user experience.

What it demonstrates

Systems thinking, AI product experimentation, API integration, automation and pragmatic optimisation.

Next step

This is a private project, so there is no public demo link at the moment. The case study shows the architecture and thinking behind it.

Build smarter.
Want to talk AI systems?

Let’s connect.

Email me  →