Inspiration
I was inspired to create this project for my sister, who is trying to expand her book and needed a space to store and view her notes more effectively. I myself have also have started dabbling in LLM usage for writing purposes, but I felt the industry UI and UX in current chatbots were inadequate for specificity - that is, referring specific memories to make the output as accurate as possible, and as such my prompts would be monolithic and extremely long, filled with text, that wouldn't be 100% accurate and require several retries to get a good enough output.
What it does
Ekrixi AI utilizes a nodic memory system, able to utilize an Gemini powered scanner to quickly synthesize information into atomic condensed notes. Notes are able to have links to other notes, which allows quick graph searching through a Zettelkasten inspired approach, similar to Obsidian and other graph/link based interfaces. Additionally, links are also descriptive, being their own defined objects with verbs and ordering inheritance enabled, allowing for a flexible structure of semantic links (Word mentions, basic relationships) to hierarchical links (folders, containers, i.e a country containing counties), and are even able to inherit note functionality to have their own links (For example, we may "Reify" the link between Country A and Country B to be "The War between Country A and Country B," and link sub notes under "The War" as battles).
The advantage to holding information in nodes, combined with a robust UI to accessing and manipulating the information manually, allows authors complete control over their universe, and able to utilize AI chatbots more intelligently, by utilizing a referencing system to refer to specific memories/notes, using similar technology that coding chatbots are able to do for referencing files.
Besides helper editor functions within the editors, the main tool to benefit from this node structure is the Story-Studio feature, which allows users to create thematic blocks and store author's notes as individual referable information as well. The UI is designed to allow a sidebar with reference to specific notes and access them, and have a dedicated chatting assistant to help write the novel, included to work with the manual text editor.
How I built it
The app is built with React with a Google Cloud Build combined with Firebase.
The schema was inspired from nodic graph structures, while including support for different view types, like hierarchical (Tree Graphs) and semantic (full mind map). I utilized inheritance for the notes to make different custom view structures for understanding the information.
Challenges I ran into
The speed of the LLM is concerning with some portions of the app. I was considering using specialized neural network approaches for certain parts of the app to make it go faster to be more cost efficient in the beginning, but it could not beat the quality of the LLM responses, outputs, and ease of implementation, leaving this as unexplored potential for a better quality scanner .
Accomplishments that I'm proud of
I was able to create a schema and system that I designed myself and implement it on an app that I'm using for myself to write and plan my own novels.
What I learned
Deploying apps through Google Cloud Run, Gemini and Key Safety, and Firebase tools.
What's next for Ekrixi AI
Major Priority: Scaling this into monetization with a pro tier for more advanced story creation Bug fixing reification and file system interactions Visual fixes on the drill down graphic interface Increase editing capabilities for stories Creating a dedicated SQL database for universes and chats to scale into the future, rather than a NoSQL approach.
Medium Priority: Fixing visual bugs for streaming AI agentic responses Creating non LLM solutions for portions of processes to minimize LLM calls and increase speed File uploads and image uploads for a more advanced chat system to collect information as text. General visual fixes across the application
Low Priority: Find a visual artist to help me illustrate a commercial and a voice over person to make me a better video
Log in or sign up for Devpost to join the conversation.