SBL (Small Brain Laboratory)

Home | Repo history | Todo | Contact

Front page

    If you enjoy making useful things, giving them away for free, and seeing people actually use them, 
    feel free to reach out! There are plenty of projects we can work on together. 
    
    Building a career while having fun with development is the best! 
    Sometimes it's better to just create something without worrying about making it perfect. 
    And if it doesn't work out? No one cares

Projects

Project Link Status Desc
(CCRE) Auto Internet URI Scrapper with Classification AI
About this
Github repo

[O][03.24.25 ~ 03.27.25] Abstract class design

[~][03.27.25 ~ ] RDS DB design by sqlalchemy library

[~][03.27.25 ~ ] DB Connect logic develpment

[~][03.28.25 ~ ] URI Scrapper logic development

[-][..] Classification AI Callback logic development

[-][..] Scoring logic development

[-][..] Test and review

[-][..] Deploy on github.io and PyPI

 [Please enable javascript engine] Concept Overview: The task is to collect all URIs (Uniform Resource Identifiers) available on the internet, identify their characteristics, and organize them. The process involves discovering all URIs, applying rules for access, and extracting information to classify and score them. The project is will be provided as a python library. Data Collection: Full text and images are not collected; instead, only the location of the text and the path to the image are collected. Full text is parsed and used to run the classification model (not saved) For images, lightweight requests like ICMP (ping) are used to verify the presence of the image, without downloading the entire image. (If you have a set image classification model, that model is used for classification) Scoring: Pages are scored based on their accessibility, which likely includes factors such as page load time, availability, and responsiveness. AI Classification: You can apply it freely by using a model that is free or paid on the Internet. It is stored and classified regardless of which classification model is used. if it returned a text. The AI classifies the content of the URIs based on the information collected. The classification is based on the content of the URIs. The content of a URI can be provided in various forms, including (text with HTML, text without HTML, request header information, binary data, and other types). You can freely transform this content and apply it to the model., Collection and save: The data is saved in a rds db(default: sqlite3) It can be changed through settings. Objective: The goal is to discover URIs, classify their content, and score them based on how accessible they are while maintaining efficient data collection.
AI TicTacToe
Training tool
Game Play [not ready]
Github repo [game]
Github repo [training]

[O][03.01.25] Preparing

[~][03.02.25 ~ ] Collecting training data (MiniMax/random)

[~][03.10.25 ~ ] Normalizing data

[~][03.12.25 ~ ] Training neural network

[~][03.09.25 ~] Developing with phaser.js (UI, effects, logic)

[-][..] Final review and deploy on github.io (tfjs)

[!][03.24.25] PAUSED

 [Please enable javascript engine] High-intelligence TicTacToe using neural network training with TensorFlow, inspired by AlphaGo. ------- [2025-03-22] The amount of data in the model is 50 million for at least one type, but not enough. I tried to get the optimal number after generating a multidimensional probability distribution with a model that lacked performance, but the computational performance was very poor. A gpu operation is required on the web to obtain numbers for at least five dimensions. I knew it would be like this before I started the project. Project progress is expected to be very slow until a good way is found.