Thomas Le

Key Features:

• AI Task Assistant: Multi-stage conversational agent using LangChain + Ollama (Qwen3:8b) with intent classification, structured tool-use pipelines, and state-machine task generation
• NLP Document Classifier: 7-category mining document classifier combining TF-IDF, Word2Vec, BERT sentence embeddings, and ensemble models (Gradient Boosting, LinearSVC)
• Automated Document Processing: OCR pipeline for 8+ file formats (PDF, DOCX, XLSX, PPTX, images) with KeyBERT keyword extraction and version-controlled storage
• Real-Time Collaboration: WebSocket group chat via Django Channels and Daphne ASGI with async consumers and persistent message history
• Custom Gantt Renderer: 860-line canvas-based Gantt chart with drag-and-drop scheduling, multi-timescale zoom, and live AJAX updates — no library dependency

My Contributions:

• Designed and implemented core Django features, database schema, task boards, and RESTful APIs
• Built the NLP document classification pipeline with domain-specific feature engineering
• Developed the document versioning system with diff tracking and audit history
• Implemented full-text search across tasks, documents, and metadata
• Authored the canvas-based Gantt chart renderer from scratch

Features:

• Unit browser with stats calculator, skills, profile, drop locations, promotion, limit break, and exclusive equipment per unit
• Enemy and equipment browsers with shareable, deep-linked modals showing stats, skills, and dagre-rendered AI behaviour graphs
• World/stage browser (chapters → zones → stages → waves/rewards/missions) and a Sanctum of Alteration floor planner (suitability/prohibition restrictions, resource gain)
• Skin viewer built on PixiJS/Spine (brotli-decompressed archives) plus a Unity WebGL viewer for fully skinned models
• Gacha simulator with pull-cost tracking and reactive community-translation layers
• Region switching (Global/KR) with automatic KR→Global data fallback

Highlights:

• Reverse-engineered bespoke binary formats (ALL4/ALLZ compression, ALTB/ALRD/ALAR stat tables, ALTX/ALIG sprite atlases) with a custom Python decoder
• Python pipeline reads the game CDN, decodes assets, and exports structured JSON + images with a one-command update workflow
• React (Vite) site with enemy/stage/unit browsing, tag pages, and a unit cost/generation simulator
• Handles multi-region data and machine-translated localisation of skill/ability descriptions

Key Features:

• Interest-based onboarding and personalised peer recommendations
• Peer Connect for discovering classmates and study partners by shared interests and communities
• Study Hub with assignment tracker, weekly planner, and peer help requests
• Campus events discovery, community groups, and support services directory
• Interactive campus map for navigating university facilities

My Contributions:

• Designed and built the full-stack application using Django with custom user model
• Implemented smart peer matching algorithm based on shared interests and community membership
• Developed study management features including assignment tracking and session planning
• Created community and support services modules with categorised filtering

Key Features:

• Role-based access control (Admin, Faculty, Student)
• Complete CRUD operations for course management
• Course assignment notification system with real-time updates
• User listing with advanced filtering capabilities
• Cross-platform compatibility (Windows, macOS, Linux)

Technical Highlights:

• Implemented CI/CD pipeline with GitHub Actions
• Deployed on AWS EC2 with PM2 process management
• Secure authentication ensuring validated user access

Features:

• Real-time game information retrieval through web scraping
• Interactive mini-games (Wordle, Word Scramble)
• Custom RESTful API queries for wiki data parsing
• Modular architecture for easy maintenance
• Deployed on Fly.io serving active communities

Contributions:

• Developed two production bots deployed on Fly.io
• Contributed to QUT Code Network's open-source bot

Features:

• Full-viewport sections with mandatory CSS scroll snapping, so each scroll gesture lands decisively on the next section
• Nested horizontal snap carousel for projects — vertical wheel input is translated to horizontal movement, passing through at the edges so the page keeps flowing
• Fixed dot navigation synced to the active section via Intersection Observer, with hover labels and keyboard support
• Top progress bar reflecting page scroll position
• Scroll-driven reveal animations using CSS animation-timeline: view() with an Intersection Observer fallback
• Multi-layer parallax hero with monotonically diverging drift rates (layers can never collide)
• Keyboard navigation: Arrow/Page keys move between sections, ←/→ move between project cards
• Dynamic project filtering, lightbox image viewer, and a scroll-to-top button
• Fully responsive with prefers-reduced-motion support and a proximity-snap fallback on short viewports

Features:

• Budget calculator based on current income and future patches
• Paid option considerations for total pull budget estimation
• Monte Carlo simulation (10,000 runs) for success rate analysis
• Multi-target unit tracking system
• Success probability visualization and statistics
• Accurate gacha mechanics with pity system implementation
• Real-time results display with success/failure breakdown

Features:

• Python + Tkinter GUI review editor for browsing and editing thousands of localisation entries with missing-entry navigation and search
• Corrections editor and rules engine that auto-fix recurring mistranslations across every file from a single rule
• Korean-asset fill for entries missing a translation, with per-region handling
• Distributes a community English patch and supports contributor pull requests

Features:

• Advanced drag-and-drop functionality
• Custom tagging and filtering system
• State management demonstration
• Responsive design with Chakra UI
• Real-time updates and animations

Features:

• Quick search functionality for F1 race results
• Modern UI with Tailwind CSS
• Responsive design for all devices
• Fast performance with Next.js optimization

Features:

• Bypass modern Android storage restrictions using Shizuku API
• Multi-language localization support
• Efficient mod file management system
• Clean and intuitive Android interface
• Secure file operations

Development Highlights:

• Level design, animations, and user interface development
• Cross-platform deployment (Android and iOS)
• Integration of third-party plugins (Google Play Services, Game Center, Facebook)
• Modular, reusable code components
• Collaboration with designers and artists

Projects:

• Single-developer game projects showcasing full development cycle
• Optimized performance for mobile platforms