bzAICallLogging-FM

Beezwax Datatools, Inc.

Previous

Next

• 1
• 2
• 3

Analyze AI Call Logs in FileMaker

Track, Analyze, and Optimize AI Interactions in FileMaker.

bzAICallLogging-FM is a demo file designed to help developers gain deeper visibility into how FileMaker’s semantic search operates under the hood. It builds on the functionality introduced in bzSemanticSearch-FM by adding AI call logging, offering insights into vector embeddings, cosine similarity scoring, and semantic search behavior.

With this demo, you can explore how semantic matches are scored, understand how vector-based comparisons are made using cosine similarity, and see how subtle changes in phrasing or formatting can influence search results. The file includes examples like comparing “hot weather” to various records and observing how slight variations in case or word choice affect the outcome.

Highlights include:
• Cosine Similarity Scoring
View how closely records match your search query using actual similarity scores.
• Verbose AI Call Logging
Enable detailed logging of AI requests to review how embeddings are processed and how thresholds impact results.
• Iterative Search Refinement
When no matches are found, the script lowers the cosine similarity threshold (up to four times) to increase the chances of finding relevant results.
• Call Log Archiving
Capture AI call logs locally and archive them in the app for easier analysis of performance and behavior.
• Embedding Inspection
Experiment with GetEmbedding, CosineSimilarity, and related functions to inspect real vector embeddings and compare them.

Whether you’re tuning similarity thresholds, evaluating the effect of word choice on search accuracy, or managing AI usage and cost, bzAICallLogging-FM offers a deeper understanding of FileMaker’s semantic search at a granular level.

This file is based on the second article in our Semantic Search blog series. Learn more about the technical background, key functions, and examples in Part 2: Key Details (https://blog.beezwax.net/filemaker-semantic-search-part-2-key-details/).