imporved UI unforently code vanished and got delted still process of fixing

AI Study Assistant — Development Log

Overview

A local AI study assistant built with Python, Flask, Hugging Face Transformers and FLAN-T5.

The project can:

• Answer questions from study material
• Generate summaries and revision notes
• Create flashcards
• Generate practice questions
• Process uploaded documents using RAG

Development

The project uses a web-based UI built with Flask, HTML, CSS and JavaScript, allowing the AI features to be accessed through a simple interface.

I also reused and improved flashcard-generation code I had previously built before Stardance existed, rather than rebuilding the feature from scratch. This earlier work became the foundation for the current AI flashcard system.

Local AI

The project uses:

google/flan-t5-base

through Hugging Face Transformers. The model runs locally, meaning the application does not require a paid AI API.

Flashcards

The flashcard system was improved by:

• Splitting documents into smaller chunks
• Generating cards from each chunk
• Parsing QUESTION: / ANSWER: output
• Removing duplicate questions
• Adding fallback handling for inconsistent AI output

Current Structure

AI Study Assistant
│
├── Flask UI
├── Document Processing
├── RAG Retrieval
├── Local FLAN-T5 AI
└── Study Tools
    ├── Q&A
    ├── Summaries
    ├── Notes
    ├── Flashcards
    └── Practice Questions

Next Step

The next stage is integrating the AI system into Frictionless, combining its AI study features with productivity, scheduling and task management.

Fixed bugs and uploaded project to Github pages

🐝 Type Bee

Type Bee is a modern, audio-first desktop spelling application built from scratch using Python and Pygame. Inspired by clean digital learning dashboards (such as assignment platforms like Satchel One), it provides an engaging typing practice tool that fetches vocabulary directly from an online 10,000 headwords PDF source.

---

🚀 Key Features

• Audio-First Challenges: Uses the pyttsx3 text-to-speech engine to read words aloud.
• Accent Customization: Easily switch between available system voice accents right from the menu.
• Online PDF Parsing: Automatically streams and extracts headwords on-the-fly using requests and pypdf.
• Satchel One-Inspired UI: Clean dashboard layout complete with dynamic button hover animations and live score tracking.
• Smooth Performance: Fully multithreaded network requests to eliminate “Not Responding” freezes on startup.

---

🛠️ Tech Stack

Component Technology Purpose Engine Pygame Window rendering, event loop, and UI graphics Audio pyttsx3 Offline text-to-speech pronunciation Network / Parsing requests & pypdf Online PDF stream fetching and text extraction Concurrency threading Non-blocking background data loading

---

📋 Changelog / Devlog

Milestone 1: Core Prototype & Audio Integration

• Initialized Pygame window loops and basic keyboard text input.
• Integrated pyttsx3 for automated word pronunciation upon round progression.

Milestone 2: Cloud Data Pipeline

• Added dynamic PDF fetching from online web sources.
• Implemented background threading (threading.Thread) to ensure the application window never locks up or reports “Not Responding” while streaming data.

Milestone 3: Dashboard UI/UX Overhaul

• Redesigned the interface to mirror clean assignment workflows.
• Added a blinking text input cursor, hover-responsive buttons, and live question/score trackers.
• Streamlined the user flow to focus purely on rapid audio spelling drills.

Updated the Sudoku game. It now allows users to play the game themself. The other option is to watch the bot solver using the algorithm described in the previous devlog. The next option is to race against the bot. The last option is to quit the game.

Tic-Tac-Toe

Overview

For my StarDance project, I decided to work on a Tic-Tac-Toe game that I originally built as part of CS50’s Introduction to Artificial Intelligence with Python.
Instead of leaving it as the original CS50 project, I wanted to develop it further and make it feel more like a proper game.
I took inspiration from the Google Doodle Tic-Tac-Toe game and attempted to recreate some of its ideas and style. I also used PySimpleGUI to create the graphical interface.

The Game

The player plays against an AI opponent:

• Player = X
• AI = O

The game has three difficulty levels:

• Easy
• Medium
• Impossible

Each level changes how the AI chooses its moves.
The Impossible difficulty uses the Minimax algorithm, which I originally learned about through CS50 AI.

Minimax

Minimax allows the AI to look at possible future moves and choose the move that gives it the best outcome.

Updated the application to include low CPU memory usage and safety checker options. Below is an image for prompt : make a picture of math student doing math on whiteboard

I have edited the code to now include options to edit the generated image. This has been neatly printed in the command line. The image below is the image generated by the AI. The prompt was to make a Pokémon character

I coded an app called Frictionless. It was an integrated modular life management platform combining AI calendar scheduling, habit tracking, Notion-style block editors, quick link folders, a gamified shopping store, loan tracking, and multi-tiered progress objectives. However, currently the app only works for me in the calendar, but by tomorrow, users will be able to set up their own integration with Google Calendar.

I have built a simple web browser with tabs that can retrieve your search results.

Ultimate Sudoku Suite

I first built a Sudoku game using Python and Pygame. Sudoku is a number-based logic puzzle played on a 9x9 grid where the goal is to fill every row, column, and 3x3 section with numbers 1 through 9 without repeating any digit.

The Rules of Sudoku

• The Grid: The board has 81 total squares divided into nine 3x3 blocks (or boxes).
• No Repeats: Each number from 1 to 9 can only appear once in every individual row, column, and 3x3 block.
• Givens: Puzzles start with some numbers already filled in as clues. Harder puzzles give you fewer starting numbers.
• No Math Needed: You do not use arithmetic or math skills; success relies entirely on logic and pattern recognition.

---

How My Solver Bot Works

To make the computer solve puzzles automatically, I built a backtracking solver.

Think of the bot like a person walking through a maze. Since it can’t see the whole maze at once, it has to remember every step it takes. My two stacks (stack_empty and stack_av) act like a trail of breadcrumbs and a list of backup choices at every fork in the road.

Here is how the logic works step-by-step:

• Taking a Step Forward:
  The bot looks for an empty spot, checks what numbers are allowed using the valid function, and picks the first legal number.

  • Stack action: It drops a breadcrumb (stack_empty) with the spot’s location, and saves its leftover backup numbers in a pocket (stack_av).

• Hitting a Dead End:
  Sometimes, the bot fills up the board, but a later spot gets completely stuck because no numbers are legal.

  • Stack action: The bot says, “Oops, this path is wrong,” and erases the failed spot back to 0.

• Backtracking (Rewinding):
  It looks at its last breadcrumb (stack_empty) and checks its backup pocket (stack_av) to see if it has any untried numbers left for that spot. If that spot is also out of options, it deletes it, erases it back to 0, and steps back even further up the stack.

Once it finds a spot that still has a backup number waiting, it plays that new number and starts moving forward again. In short, the stacks let the bot undo its mistakes one step at a time instead of starting over from scratch!

TradeCraft is officially up and running! The Flask-powered stock market simulator now features secure OTP email authentication, real-time live pricing via yfinance, an interactive trading engine, and dynamic portfolio tracking—giving users a complete, hands-on investing experience from scratch.