AI

KGeN AI

We have explored AI in a few areas within KGeN and we plan to put some of these ideas in production in 2025.

Avatar generation

For every PoG card that gamers mint, they get a customized gamified avatar based on their selfie and the parameters they select. We have used a custom face swap model(models used: Llama/Claude, Instant ID/ flux and Lora models) working in collaboration with segmind to generate this Avatar.

The gamer selects genre, character, style and selfie as inputs.

Genre is 1 of the following:

• Adventure

• Fantasy

• Action

• Toon

• Sci-Fi

• Horror

Characters depend on the genre. Following are few examples:

• Action - Soldier

• Adventure - Explorer

• Fantasy - Elf, Mage

• Scifi - cyborg

• Toon - Queen/King

• Horror - Zombie, Vampire

The style is common across genres and characters.

Exploratory projects

Following AI projects are in exploratory phase:

PoG score calculation using AI:

At this point, we are calculating the PoG score using different gamer attributes and the final PoG score is calculated using a formula based on different aggregated parameters. We plan to enhance this using AI based approaches. AI based approaches need training data and for that we need a way to generate scores for users without using formulas. Using training data with formulas would result in overfitting. The training data can be done by experts rating the score or simulating variations in the formula based data. AI based approach reduces bias and it works well when the scale of input parameters are high. The plan is to do reinforcement learning.

Gamer recommendations:

Provide gamer engagement recommendations based on their previous interests using Collaborative filtering/Content based filtering.

Gamer risk score analysis:

Profile gamer risk score and pay the rewards in advance to reduce rewards TAT.

Internal/External assistants:

Use a combination of LLM and RAG on top of the KGeN database to address internal and external queries.