if memory2 query returns a graph, how can we present it to a LLM model in a way in which it understands it?
write an eval, ask basic questions like
"what is the time range of this graph"
"where is the lowest point"
"what's the general shape of the graph"
"do you see any peaks, and at what points in time?"
Find a format that allows agent to answer these things without running aditional analysis on the data.
But also potentially allow model to do aditional analysis and see which formats work best.
As an experiment, we can ask real questions from an agent, like "describe the scene before you entered dark room to the south"
"how many times have you passed the couch in the middle of the office"
if memory2 query returns a graph, how can we present it to a LLM model in a way in which it understands it?
write an eval, ask basic questions like
"what is the time range of this graph"
"where is the lowest point"
"what's the general shape of the graph"
"do you see any peaks, and at what points in time?"
Find a format that allows agent to answer these things without running aditional analysis on the data.
But also potentially allow model to do aditional analysis and see which formats work best.
As an experiment, we can ask real questions from an agent, like "describe the scene before you entered dark room to the south"
"how many times have you passed the couch in the middle of the office"