A Tale of Two Heroes: Azure Form Recognizer vs. AI Builder

#powerfuldevs #azure #powerplatform #ai

Intro:
Batman and Azrael are two iconic characters from the DC Comics universe, each with their unique approaches to crime-fighting and justice.

Testing the Limits of Heroic Vigilance:
In our crime-fighting comparison, we staged a showdown between the iconic Batman and his successor Azrael. To challenge their abilities, we introduced a mix of challenging scenarios, analogous to invoices with various noise elements like high brightness, Gaussian blur, mirror flip, blurred text, and skewed layouts. Both heroes were then put to the test, just as we evaluated the responses generated by Azure Form Recognizer and AI Builder. We carefully analyzed the key elements of their performance, scrutinizing the data attributes and confidence levels of their respective crime-fighting techniques, shedding light on the efficacy of these two champions in their battle against the criminal underworld.

Gameplay Review: What We Learned from Playing:

1) The initial finding revealed that Azure Form Recognizer consistently produced more reliable key-value pair output

2) Azure Form Recognizer proved to be efficient in capturing table-level information

3) In the case of Azure Form Recognizer, the confidence levels for each key-value pair consistently outperformed those generated by AI Builder

4) Both models encountered difficulties when processing invoices with noise, such as high brightness and mirror flip, as they failed to capture these elements

5) Azure Form Recognizer effectively processed invoices with skewed orientations, whereas AI Builder primarily provided basic metadata for key-value pairs. Additionally, the confidence interval in Azure Form Recognizer surpassed that of AI Builder

Test Results Table

		AI Builder		Azure Form Recogonizer
Type	# Invoice	Avg Keys Identified	Avg Confidence Interval	Avg Keys Identified	Avg Confidence Interval
Orignial Invoice	20	11	0.738	13	0.81
Brightness Skewed	5	8	0.7	13	0.79
Gaussianblur	5	2	0.08	4	0.12
mirror flip ( AFR read the text in mirror and captured in tables)	5	1	0	1	0
skewed orientations	5	10	0.694	13	0.81

Batman and Azrael each bring their unique strengths and approaches to crime-fighting, decision-making in the realm of data processing and AI tools requires a multi-dimensional analysis and below is a starting point that I would recommend

Much like Batman's strength lies in his ability to adapt and overcome challenges through intelligence and resourcefulness, Azure Form Recognizer's out-of-the-box Invoice pre-built model has demonstrated superiority when handling noise and delivering consistent output. In contrast, just as Azrael may seem more powerful and straightforward to use, AI Builder tends to struggle with consistency.