Sign In

James Croft

Microsoft MVP & Principal Software Engineer at Razor Ltd

James Croft is a Principal Developer at Razor with over a decade of experience in software development, specializing in creating innovative user experiences across various platforms including Windows, Android, iOS, and the web.

He holds a Bachelor of Science in Computer Science with Industrial Experience, achieving a 1st at The University of Hull, and has a background in technology spanning multiple industries such as customer service, construction, manufacturing, public service, government, health, and social care.

As a Microsoft MVP in Windows Development, James is distinguished for his contributions to the technical community, which include open-source projects, mentorship, publications, and speaking engagements at conferences and user groups.

His expertise extends to emerging technologies like mixed reality (MR/AR/VR), wearable devices, IoT, and conversational AI, showcasing his passion for the ever-evolving landscape of technology.

Prior to his current role at Razor, James held various positions at organizations such as Black Marble, Seed Software, Platform Expos, The Hullfire, and Jkc Real Estate LLC, demonstrating a diverse and rich professional background in software development and technology leadership.

Highlights

Aug 29 · linkedin

Looking to evaluate SLMs and LLMs for structured JSON data extraction from both structured and unstructured documents?

I've updated my article from earlier this year with new evaluations for GPT-4o Mini and Phi-3.5 Mini models to give you an indication on their performance for these use cases!

https://lnkd.in/ezNsvapa

Here's what you can expect to uncover:

📃 Accuracy & Speed - GPT-4o using Vision capabilities still excels in both accuracy and speed for extracting data from both complex structured and unstructured documents. 📃 Cost Effectiveness - For high-accuracy requirements, GPT-4o using Vision also provides the best choice for keeping costs low for document processing. 📃 SLMs - For simple text based documents, Phi-3 Mini can be considered. However, they struggle with complex domain specific language without more detailed extraction prompts. 📃 Structured Documents - All models provide high accuracy for extracting data from structured documents, such as invoices. However, there are key differences in speed and cost which must be evaluated.

#Azure #DataExtraction #OpenAI #DocumentIntelligence

Aug 5 · linkedin

I share a lot of learnings about adopting Azure AI Services for extracting structure data from documents, and now you can get all of these insights in one place.

Get the resources in our Learn collection 👉 https://lnkd.in/eb6R-2Zy

This handy link brings all of our articles and GitHub samples into a single resource that you can bookmark and come back to!

In this resource, you'll discover:

💡 Combining Azure AI Document Intelligence and Azure OpenAI to extract structured Markdown and JSON from documents 💡 Evaluating document data extraction prompts for both SLMs and LLMs in a secure Azure AI Studio environment with Prompt Flow 💡 Using GPT-4o's vision capabilities to extract insights from document images

And we'll keep updating this collection as we explore more concepts in this space!

#Azure #AI #Learning

James Croft
Add to my network