What are the key considerations when designing a context graph for AI applications?

Context graphs are essential for AI applications as they help in understanding relationships and semantics. However, each AI company may require a unique context graph tailored to its specific needs. What factors should be considered when designing these graphs? How do you ensure they are flexible and scalable? Share your thoughts and experiences!