Eny Who?
That's me!
You can find me always keeping on top of the latest online trends, scrolling through feeds, reading obscure articles, watching YouTube videos, and chasing after my dog.
If you want strategy or content creation that brings authenticity to your brand, reach out and let's make some cool stuff!
Other Projects
This page is currently a work in progress but I'm excited to share the interesting personal projects I've been working on.
Re:ZERO AI Knowledge Base
Background
I'm a bit of an anime nerd, among other things. Re:ZERO - Starting Life in Another World by Tappei Nagatsuki is one of my favorite pieces of fiction because of the deep complex lore.
There are currently* 44 Japanese Light Novels, 27 of which has been officially translated to English, 769 Web Novel chapters, 354 Side Stories, 66 Anime Episodes, 66 Anime Shorts, plus hundreds of author insights via social media posts and Q&As (not found in any other form).
How is someone expected to remember all the details?
What This Is
I am creating a structured knowledge base that is the foundational component for an AI, specifically an LLM, that can answer questions about the Re:ZERO universe. I found current LLMs unable to answer accurately due to their general purpose use.
This knowledge base, in its current state, is comprised of over 2 million chunks of data (text broken down into smaller manageable sized segments that is then retrieved by the LLM) from all the written source materials (Light Novels, Web Novels, Side Stories, etc). This data is organized so the AI can find exactly what it needs and cite where it came from.
The idea of this knowledge base is to be an unbiased and factual source of reference material from canonical sources. This level of data allows someone to ask complex abstract questions and receive deep, insightful, and factual responses.
Current State
Currently, the AI is able to answer most questions with detail and high accuracy. It sometimes can struggle retrieving all the relevant material to provide a comprehensive answer but this is something that I've been continuously improving through system prompt engineering (refining the AI searches and responses) and improvements in data retrieval.
The finished state would be when 90-99% of questions can be answered accurately with sources provided and credit when applicable (in the case of unofficial translations).
AI Knowledge Base Builder
Background
Creating my Re:ZERO knowledge base took over a year of research and work to get a satisfying working version. There must be an easier way for everyone to do what I did in a shorter amount of time.
There are many communities that would benefit from having a super-powered upgrade to fan wikis, fostering deeper discussions within fandoms without needing to learn programming.
What This Is
I'm creating a desktop application for building high-quality, searchable, citation-aware AI knowledge bases for fandoms. This application aims to bridge the gap between technical capability and user accessibility, feeling sophisticated yet approachable for nontechnical users. Designed so ordinary fans can create wiki knowledge bases and AI companions for their favorite universes and stories.
Users upload text based content and can either let the application build the database directly, or refine their dataset manually for more accurate responses. The entire process is completely human-readable. No code required.
Current State
I have created a minimal proof-of-concept application that can currently process documents and utilize in-app AI to test retrieval accuracy. I'm currently implementing core functionality, including a modular metadata editor.