LLM-Lab
July 2026LLM / AI
LLM Lab is an interactive platform I built while learning about Large Language Models through the Hugging Face LLM course. I noticed that many learners understand the theory but struggle to visualize what actually happens inside an LLM. So, instead of building another chatbot, I created a learning-focused environment where users can experiment with generation parameters, compare models, and visually follow the journey from text → tokens → token IDs → tensors → generated output. The platform is designed around a simple idea: if you can interact with it, it becomes easier to understand.
An interactive learning platform that turns the LLM inference process from a black box into something students can see, experiment with, and understand.
Tech Stack
NextJSTypeScriptTailwind
Features
- Interactive playground for experimenting with LLM generation
- Real-time visualization of tokenization and inference steps
- Temperature, Top-K, Top-P, and repetition penalty controls
- Side-by-side model comparison
- Inference visualizer showing tokens, IDs, tensors, padding, and attention masks
- Learning section covering core LLM concepts
Challenges
- Understanding the inference pipeline deeply enough to visualize it correctly
- Turning complex LLM concepts into simple interactive explanations
- Handling API responses and keeping the interface responsive
- Designing visualizations that are informative without becoming confusing
Learnings
- Gained a much deeper understanding of how LLM inference works
- Learned how tokenization, sampling, and model parameters affect generation
- Improved my understanding of Hugging Face APIs and model interaction
- Learned how to turn technical concepts into interactive learning experiences
- Improved my Next.js and TypeScript development skills
Highlights
- Built as a learning project, LLM Lab is now a live educational platform that I plan to continuously expand with more visualizations, experiments, and LLM concepts