Founding Engineer @ Collate | Building OpenMetadata
Markdown supported · ⌘↵ to post
The tiered result is the useful part: a context layer is not a yes-or-no buy, it is a question of where the agent is guessing. For simple questions the agent already knows the table, and extra context is mostly cost. For the hard ones — joins, naming that does not match the question, ownership a model cannot see — the context is the difference between a wrong answer and a right one. I see the same shape building Homeowner HQ for realtors and lenders, launching December 4 on Amazon AWS: a question about a home needs the right record, the right loan, and the right person, and a wrong join there is not a metric, it is a homeowner given the wrong answer. Thanks for measuring calls and tokens, not just accuracy.
Steven Mock
The tiered result is the useful part: a context layer is not a yes-or-no buy, it is a question of where the agent is guessing. For simple questions the agent already knows the table, and extra context is mostly cost. For the hard ones — joins, naming that does not match the question, ownership a model cannot see — the context is the difference between a wrong answer and a right one. I see the same shape building Homeowner HQ for realtors and lenders, launching December 4 on Amazon AWS: a question about a home needs the right record, the right loan, and the right person, and a wrong join there is not a metric, it is a homeowner given the wrong answer. Thanks for measuring calls and tokens, not just accuracy.