Taking the output of many dumb LLM search relevance judges and feeding the output to a decision tree to improve precision
Expanding on my previous post, I show the impact of checking both directions in pairwise evaluation
Add friction to your twitter login to keep yourself sane.
I built AI-bot twitter and learned they like to argue about pumpkin spice lattes
The book AI Powered Search is out and I couldn't be more grateful to Trey Grainger and Max Irwin for having me on this journey
If you have discipline to throw away your first idea, draft, throwaway PRs often drives more progress than a design doc.
Some notes when you get into Go from Python for the fellow Go newb.
Large companies can't put the genie back in the bottle because to most employees, they don't have autonomy over their roles
Elasticsearch doesn't have a straight-forward way to match the 'full' field (all the tokens as a phrase).
Reciprocal Rank Fusion, while a useful tool, doesnt magically make hybrid search relevant
A notebook showing the real decisions computing search evaluation stats
The lack of objective definition of good search creates huge hazards when creating search, RAG, AI solutions
Avoiding conflict is the death knell of organizations that leads to a lack of progress and careers that implode.
In reality, staff engineers aren't about 'company wide' impact but a system of patronage where managers reward behaviors they value