Flynn Cruse | Software & AI Portfolio

Tutechy-Frontend

Description: A Next.js-based frontend for the Tutechy educational platform offering separate interfaces for students, teachers, and tutors with advanced learning features.

Key Features:

• Role-based access control with separate interfaces
• AI-powered tutoring with natural language processing
• Real-time chat with mathematical notation support
• Voice input with speech-to-text capabilities
• Curriculum management tools for teachers
• Progress tracking and analytics

Technology Stack: Next.js 15.1.0 framework, TypeScript, AWS Amplify/Cognito, Redux Toolkit, Tailwind CSS with shadcn/ui, KaTeX and MathJax for math rendering

Ed-Tech-Platform-DB-API

Description: A Node.js API service providing database access and business logic for the Tutechy educational platform, connecting to PostgreSQL using Prisma ORM.

Key Features:

• Complete API for educational platform backend
• Role-based user management (students, teachers, tutors)
• School and curriculum management
• Security middleware with rate limiting and input sanitization
• Detailed monitoring and logging

Technology Stack: Node.js with TypeScript, Express for API routing, Prisma ORM for database interactions, PostgreSQL database

Cletude-Prompt-Service

Description: A FastAPI-based service for managing prompt generation and processing for the Cletude/Tutechy educational platform, supporting flexible LLM provider integration.

Key Features:

• Robust API for prompt management, generation, and storage
• Flexible architecture supporting multiple LLM providers
• Vector storage using PostgreSQL with pgvector extension
• Session management for contextual conversations
• Streaming response capabilities for real-time interactions

Technology Stack: Python 3.10+, FastAPI framework, PostgreSQL with pgvector, Docker containerization, Poetry for dependency management

Wordle

Description: A Unity implementation of the popular word-guessing game Wordle, where players have six attempts to guess a five-letter word with feedback on letter placement. Part of the TuTechy educational platform.

Key Features:

• Complete Wordle game mechanics in Unity
• Educational project for learning game development
• Available as open-source tutorial
• LLM integration generating subject-specific questions based on teacher-uploaded curriculum
• Integration with TuTechy educational platform for student engagement

Technology Stack: Unity 2021.3 (LTS), LLM integration, TuTechy platform APIs

Slither.io Education Version

Description: An educational adaptation of the popular Slither.io game, modified for classroom use with learning components integrated into gameplay. Part of the TuTechy educational platform.

Components:

• Unity-based game implementation
• Educational modifications to the original Slither.io concept
• Comprehensive documentation
• LLM-powered question generation system allowing students to earn retry/extra-lives
• Integration with TuTechy platform for classroom deployment

Technology Stack: Unity, LLM integration, TuTechy platform APIs

TD3D-UnityGame (Tower Defence 3D)

Description: A 3D tower defense game with 20 levels featuring strategic tower placement, enemy waves, and progressive difficulty.

Key Features:

• 20 unique levels with increasing difficulty
• Star-based reward system based on castle health
• 16+ unique tower upgrades with different capabilities
• 9 unique enemy types with varying stats and resistances
• 4 special items to assist gameplay
• Mobile, web, and PC compatibility
• Educational component using LLMs to quiz players after each wave on study topics
• Reward system that grants additional in-game currency for correct answers, enabling better tower purchases/upgrades

Technology Stack: Unity, C#, Binary serialization for save data, LLM integration for educational component

Model Walkers

Description: An innovative turn-based AI Video Generation powered card game where players create custom 'cards' using natural language. A large language model (LLM) that balances gameplay guides the game by calculating appropriate mana costs for each card; arbitrating between the player's desired effects and the game's balance mechanics. The game features visual generation through Luma AI integration, creating dynamic video representations of card effects.

I built this game in approximately 2 hours during a company retreat as a rapid prototype demonstration!

Key Features:

• Players start with 1 mana, gaining 1 max mana each turn (up to 10)
• Interactive card creation system where players describe desired effects in natural language
• Dynamic balancing through LLM-calculated mana costs based on card power
• Visual representation of cards and effects using Luma AI's video generation API
• Battlefield interaction with creatures and enchantments
• Graveyard system for defeated creatures and used spells

Technology Stack: Python 3.10+, Pygame 2.1.2, LLM integration, Luma AI API for video generation

MCTS-LLM-Swarm Research

Description: An AI reasoning framework integrating Monte Carlo Tree Search (MCTS) with Large Language Models (LLMs) to systematically explore solution paths for complex problems.

Key Features:

• Asymmetric tree exploration algorithms for complex problem solution spaces
• Balanced exploitation of promising reasoning paths with exploration of diverse alternatives
• Performance scaling with computational resources at inference time
• Systematic search methodology that improves with additional compute allocation

Technology Stack: Python, OpenAI API, Jupyter Notebook

Neurology Assistant

Description: An experimental OpenAI-based AI assistant project focused on neurology, leveraging RAG (Retrieval Augmented Generation) on MIT neurology lecture material.

Key Components:

• RAG training methodology using neurology lecture content
• Integration with OpenAI's Assistants API
• Structured conversation threads and message handling

Technology Stack: Python, OpenAI Assistants API

Self-Corrective Code Generation

Description: A multi-model code generation system using multiple LLMs in a collaborative workflow to produce high-quality, working code through iterative improvement.

Features:

• Integrated multiple models including Mistral AI, IBM Granite, and OpenAI GPT-4
• Developed a LangGraph-powered workflow for collaborative code generation
• System autonomously executes, validates, identifies issues, and iteratively improves code
• Focused on achieving working solutions by combining strengths of multiple models
• Implemented high-performance XML parsing as a practical application

Technology Stack: Python, LangGraph, Multiple LLM APIs, Testing frameworks

AIAAR (AI Assistant for Auto Recording)

Description: Chrome extension for Google Meet that automatically records, transcribes, and analyzes meeting content with AI assistant capabilities.

Key Features:

• Meeting transcript capture from Google Meet captions
• Private, browser-based processing without server uploads
• AI-powered meeting analysis and query functionality
• Automatic transcript export at meeting end

Technology Stack: JavaScript, Chrome Extension APIs, AI integration APIs

RAG MCP Tooling

Description: A Retrieval-Augmented Generation (RAG) system using Model Control Protocol (MCP) for machine learning and deep learning concept retrieval.

Key Features:

• Transformed "Fast AI for Coders With FastAI and PyTorch" resource into a vector database
• Built a complete RAG pipeline for machine learning and deep learning concepts with code examples
• Integrated with MCP for seamless retrieval within agent interfaces
• Focused on improving agentic code generation performance for ML/AI projects

Technology Stack: Python, Vector databases, MCP integration

Groq-Verbal-Assistant / Groq-Speech-Assistant

Description: Voice assistant implementation leveraging Groq's high-performance LLM inference platform for faster response times and natural voice interactions.

Features:

• Speech-to-text conversion using Deepgram API
• Groq API for low-latency LLM responses
• Text-to-speech for verbal output
• Web interface for voice interaction
• Customizable AI personality through prompt modification

Technology Stack: Python, Flask, Groq API, Deepgram API, Web Audio API

Startup-Financial-Forecaster / Projections

Description: Financial modeling application for startups with comprehensive forecasting capabilities for business planning, cash flow management, and growth scenarios.

Features:

• Employee cost modeling with salary progression tracking
• Fundraising planning and round simulation
• Revenue stream modeling with multiple business models
• Expense management and categorization
• Monte Carlo simulation for risk analysis
• Equity management and vesting schedules

Technology Stack: Python, Streamlit, SQLite, Plotly

AI Compliance Navigator

Description: A Streamlit-based application for querying multiple specialized OpenAI Assistants trained on specific compliance documents with individual and aggregated responses.

Key Features:

• Multi-select interface for choosing compliance documents
• Real-time progress tracking for query processing
• Individual response tabs for each selected assistant
• Aggregated response synthesis for multiple selections

Technology Stack: Python, Streamlit, OpenAI Assistants API

RayBans Meta Glasses Augmentation

Description: A computer vision system designed to work with Ray-Ban Meta AI glasses, inspired by Harvard students Nguyen and Ardayfio's I-XRAY project. The project paired OBS Virtual camera with facial detection for RayBans Meta glasses livestream.

Key Features:

• Integration with OBS Studio for Instagram livestream capture
• Real-time facial detection using TensorFlow-based detection models
• Project ceased after facial detection was proven due to ethical concerns
• Planned extensions included PimEyes API for face identification, Twilio for communication, and database implementation

Technology Stack: OpenCV for video processing, TensorFlow for facial detection, Python, OBS Studio