Day 1


🧠 Transformer Model Text Analysis

We implemented a transformer-based large language model (LLM) to analyze text and conversational data.
This project provided hands-on experience with the architecture and inner workings of transformer models,
while deepening our understanding of how LLMs process, interpret, and generate natural language.

The goal:
To explore how transformer models operate and to gain practical experience in building, training, and applying an LLM to real-world text inputs.


🔗 Attention Mechanism

The image below shows an attention map, which visualizes how correlated each word is to the others in a sentence.
This is a core mechanism in transformer models—it allows the model to weigh the importance of each word relative to every other word.

Attention Map


💬 Conversation Summarization

Below is an example of how the model was trained using real conversations.
It shows how the transformer learns to summarize long text inputs into concise outputs.

Conversation Summarization Training

Day 2


📊 Pokémon Stat Embedding Visualization

🧬 About the Plot

We created embeddings for Pokémon based on core stats such as HP, Attack, Defense, and Speed. This data was pulled using a public Pokémon API and visualized using a dimensionality reduction technique called t-SNE.

Unlike other projection methods that may impose artificial structure, t-SNE preserves the local relationships in the data, giving us a more authentic view of how Pokémon compare based on their attributes.

Each point in the plot represents a single Pokémon, and you can hover over each point to view its types, abilities, and evolution line.

Day 3


🌐 Pokémon Network Visualizations


🧬 Evolution Network


🏞️ Habitat Network (Community Detection)

📉 Pokémon t-SNE Plot (Colored by Habitat Communities)

This t-SNE visualization displays Pokémon as nodes positioned based on their stat-based similarity, with colors representing their habitats. Each point is a Pokémon, and nearby points tend to have similar stats like HP, Attack, Defense, and Speed.

🌍 Observations

This visualization offers a high-level view of how Pokémon types and habitats can influence their stat-based groupings.