KTL Research Hub
Kitty and Tweak's research and development universe
Independent AI and machine learning research, language model architectures, LLM training and fine-tuning, tokenisation, reasoning models, neural networks, dataset engineering, model evaluation and benchmarking, AMD ROCm and NVIDIA CUDA acceleration, efficient inference, AI agents, cybersecurity, distributed computing, open-source development and reproducible experimental research.
Purple cyber-neko imagination × crimson workshop defiance.
Independent AI research, systems engineering, cybersecurity and experimental software.
Explore the Lab · Our Repositories · Hugging Face
We’re KrazyKitty and King Tweak — two independent builders with different instincts, a shared love of technology, and far too many questions to leave the defaults alone.
One universe glows violet: cyber-neko character, connected systems, Python, security and a terminal full of ideas. The other burns crimson: an obsidian workshop, experimental model architectures, Linux and the urge to take a system apart until it makes sense.
KITTY × TWEAK LABS is where those worlds meet. We turn curiosity into code, challenge each other’s assumptions, and work towards experiments someone else can reproduce.
The Cyber-Neko Systems Architect
Ginger hair, glasses, cat ears and a purple hoodie: the character at the centre of an interconnected developer universe. Violet terminals and playful details sit alongside a serious interest in how software, security, infrastructure and AI fit together.
Kitty’s instinct is to follow the connections — from a Python tool to an automation service, from a home server to a research environment, from a language model to the data and evaluation that make its results meaningful.
Her territory: Python · cybersecurity · AI/ML · Linux · systems architecture · HomeLab · automation.
Her independent universe: KittyLM for language-model research, KERNEL//KITTY for HomeLab infrastructure and observability, and KITTY//LINUX as her planned Linux distribution.
Lavender, electric violet and orchid. A little feline mischief. An entire universe to understand and build.
GitHub · Hugging Face · Website
The Crimson Workshop Engineer
The Blackblade archetype: dark-fantasy attitude, underground developer energy and a workshop built around understanding the machinery. Obsidian surfaces, deep crimson and industrial details give his world its own unmistakable character.
Tweak’s instinct is to open the model, question the architecture and experiment with what could work differently — from tokenisation and memory to inference, optimisation and the tools surrounding the system.
His territory: language-model architecture · tokenisation · model engineering · NVIDIA/CUDA · Linux · developer tools · performance experimentation.
His independent universe: TweakTron and its model experiments, alongside TweakOS and his Linux development environment.
Crimson, obsidian and a rebellious streak. Every experiment is another reason to get back to the workbench.
GitHub · Hugging Face · Website
Our interests overlap across AI, systems and tools. We each bring engineering judgement, creativity and independent projects to the table.
| Shared research space | Questions we want to explore |
|---|---|
| Models & architectures | What changes a model’s behaviour, efficiency and capabilities? |
| Data & tokenisers | How do provenance, data quality and vocabulary choices affect results? |
| AMD × NVIDIA | What transfers between ROCm and CUDA, and what needs different treatment? |
| Systems & security | How do we build environments we can understand, observe and protect? |
| Agents & tools | How can useful automation remain inspectable and permission-aware? |
| Evaluation | What can we measure, reproduce and explain? |
| Programme | Purpose | Planning stage |
|---|---|---|
| KTL-020 // KTL Bench | Reproducible model and hardware evaluation | Initial implementation candidate |
| KTL-006 // CrossSilicon Lab | AMD/ROCm × NVIDIA/CUDA comparison | Initial implementation candidate |
| KTL-003 // K×T DataForge | Dataset validation, provenance and preparation | Planned research |
| KTL-002 // DualForge | Controlled experiments across compute backends | Planned research |
| KTL-014 // Compute Fabric | Secure research jobs and compute coordination | Future development |
| KTL-001 // KittyTron | Future joint model research informed by both model families | Concept / feasibility research |
These are research directions and planning stages. They do not imply completed implementations or validated model capabilities.
Explore Kitty’s public research, Tweak’s public research and the KTL organisation. Each repository’s model card, dataset card or Space documentation is the place to check its purpose, licence, limitations and release state.
Independent development projects mentioned here are part of our wider ecosystems; they are not automatically published Hugging Face artifacts. A reference or collection entry credits an original source and does not make it KTL-owned.
KittyLM remains KittyLM. TweakTron remains TweakTron. Independent work keeps its identity and attribution. Joint research is identified explicitly, and third-party work is credited to its original authors.
Curiosity gives us the idea. Experimentation tests it. Evidence determines what we can claim.
We keep Verified, Reported, Inferred, Proposed and Unknown distinct. Public availability, research maturity and supporting evidence are recorded separately.
Our planning framework is the KTL Master Project & Research Overlay v1.2 — Engineering Execution Edition. Publication follows the relevant validation, licensing and release review.
Our compute environments remain independently administered. Shared experiments have explicit scope, attribution and access boundaries.
Enter the research repository →
KRAZYKITTY 💜 CODE. CYBER. CHAOS.
KING TWEAK ❤️ BUILD. BREAK. TWEAK.
Built together, with curiosity, care and a shared refusal to stop learning.