aipwadeep-techproduct

Live

1 day

Time to first working version

230 iterations over 5 days to reach flow state

Meridium

A custom, AI-native alternative to Obsidian that stores knowledge as an interconnected synaptic network — with a Supabase backend and a Claude-powered chatbot that navigates your notes for you.

Try it

Meridium reimagines the personal knowledge vault. We loved the concept behind Obsidian — vaults, linked ideas, a second brain — but plain text files felt archaic, and the entry experience was heavier than it needed to be. We wanted something that felt like a modern note-taking app: fast, frictionless capture paired with an AI that could actually reason across everything we'd stored.

Instead of loose files, Meridium models knowledge the way the brain does. Entries form synaptic connections to one another, creating a web of related information rather than an isolated pile of documents. That structure isn't just conceptual — it makes the data dramatically more useful to an AI. A connected knowledge graph gives the model rich, relevant context, so answers are grounded in what you actually know.

The system runs on a Supabase backend with a direct API connection to Claude. On top of the core note-taking layer we added task management, gamification, analytics, a content management layer for your data, and a live visualization of the connections between entries. The result is a self-hosted tool that replaces a subscription we'd otherwise pay for — built in days, and tuned to fit exactly how we think.

Screenshots

Features

What it does

Full-featured note taking

Everything you'd expect from a modern note-taking app, with capture designed to be fast and frictionless.

AI auto-categorization and tagging

New entries are automatically classified and tagged by AI, so organization happens without manual upkeep.

AI writer

An assistant that helps fill in gaps and flesh out entries as you write.

Lumen chatbot

A Claude-powered chatbot with access to all your entries as knowledge, able to navigate the synaptic network and answer questions grounded in your own data.

Synaptic knowledge connections

Entries link to one another the way neurons connect, mirroring how the brain stores and retrieves information.

Synapse visualization

An animated, visual representation of the entire connected knowledge graph.

Task tracking and gamification

A built-in to-do tracker plus metrics and gamification, including streak tracking, to keep engagement high.

RSS capture pipeline

Scans multiple RSS feeds from reputable media sources and surfaces relevant articles that can be captured, classified, and folded into your connected knowledge base — with full import and export support.

Process

How we built it

Audited Obsidian

Started by studying Obsidian closely, cataloging what worked well and what felt archaic or high-friction. That teardown became the requirements list for what Meridium needed to keep, fix, and add.

Prototyped the interface

Used our own design tool to build a working prototype, focusing on making note entry effortless and immediate.

Designed the data model

Built out the backend schema in Supabase and expanded the scope well beyond notes — adding task management, gamification, analytics, a data CMS, and a visualization layer for the connections between entries.

Became our own usability testers

Started using Meridium daily as its primary users, surfacing friction points and dead ends through real, everyday use rather than hypothetical scenarios.

Iterated to a stable, locked version

Cycled through 230 versions over five days, refining the architecture and the experience until the final build was locked in.

Reflections

What we took away from this project.

The complexity curve

This was the most sophisticated thing we've ever built, and the complexity was off the charts. We ended up rebuilding it from scratch three times before landing on a stable architecture. The hard lesson: adding major features on the fly without planning for them consistently introduced stability problems.

Synapse and Lumen delivered

Synapse, the visual display of the entire connected dataset, came together in a single prompt. And once a meaningful amount of data was entered, the Lumen chatbot proved able to answer questions and produce results that were far more relevant and accurate than ChatGPT or Claude could achieve on their own — because it was reasoning over a purpose-built, connected knowledge graph.

Flow state is earned in the details

The biggest takeaway was about achieving flow state — the point where an app is so good at its task that the act itself becomes second nature. It means zero friction, no dead ends, the least possible user input for the maximum result. Reaching it took days of testing, because the wins come from small, deliberate refinements: leveraging progressive disclosure to strip anything that adds noise to primary tasks and hiding exceptions until they're actually needed.

Want something like this for your business?

Tell us what you're working on. We'll tell you how fast we can build it.