mathieuisabel

[mathieuisabel]

Comparing Jobs Using Semantic Similarity To Support Decision Making

How do you effectively compare items that appear distinct on the surface, like tasks, financial products, or even claims? Semantic similarity offers a way to assess and measure these differences quantitatively, enabling better decision-making across a variety of domains. Understanding Semantic Similarity At its core, semantic similarity transforms item descriptors (or attributes) into a vector…

[mathieuisabel]

Content Analyzer - Audio Transcription Support

Today, we’ll explore a recent enhancement to the Content Analyzer: audio file analysis. This new capability allows businesses to uncover powerful insights from voice interactions by transcribing and analyzing audio files. In this blog post, we’ll showcase how the Content Analyzer works with a real-world example—a contact center call recording. Imagine a scenario where a customer calls to cancel…

[mathieuisabel]

Navigating Grey Zones - Picking an ETF

Introduction Exchange-Traded Funds (ETFs) have become a cornerstone in many investment portfolios, offering the benefits of diversification, liquidity, and cost-efficiency. However, with thousands of ETFs available, choosing the right one can feel overwhelming. How do you balance factors like fees, asset composition, historical performance, and risk? The answer lies in using data-driven tools to…

[mathieuisabel]

Enhancing User Inquiry Responses: A New Multi-Task Planning Approach

In the realm of conversational AI, user inquiries can often be high-level and complex, making it challenging to provide high-quality answers with a simple Retrieval Augmentation Generation (RAG) approach. To address this, I’ve worked on developing new capabilities that break down the inquiry process into detailed, actionable steps, ensuring thorough and contextually accurate responses. Tackling…

[mathieuisabel]

Improving Unstructured Data Extraction Capabilities

While continuing to improve the capabilities of what can be achieved with the LLM Planner and Executor agents, a few updates were made to the extraction capabilities in the platform to provide more robust and detailed data extraction from unstructured content, streamlining the process and enhancing the accuracy of the extracted information. Let’s go through some of the enhancements. Support for…

[mathieuisabel]

LLM Based Planner: Better Plans Through Better Objectives

As I continue to explore what’s needed to build a robust LLM based planning agent, one thing I’ve been focusing on lately was how to improve the quality of the plan itself. One dimension I’ve been exploring is how refining the objective of the job and tasks themselves to provide a clearer and more adjusted plan. I’ll go over some of the things that provided improvements in that area. Improving…

[mathieuisabel]

LLM-Based Job Planner and Executor Agents For Task Management and Execution

While working on research and analysis capabilities for my personal project, I started to dive into the possibility of automating further the generation of what needs to be done in order to reach a particular objective. In the current iteration of the Content Analyzer (see Content Analyzer posts), you articulate the objective in terms of what you want to accomplish with the content but I wanted…

[mathieuisabel]

Content Analyzer Updates - Content Q&A Support and More...

Another week, another post! :-D

Content Analyzer has been augmented with a suite of new features designed to streamline the process of content analysis, especially when dealing with unstructured data. These enhancements include support for PDF content ingestion, the introduction of criteria under analysis aspects, advanced data extraction capabilities, question answering based on content analysis, and generating proposals based…

[mathieuisabel]

Content Analyzer - Building on top of Dynamic Structuring of Content Analysis

In my previous post, I went to some level of technical details about core API capabilities that were built to support a project I’m working on. As the continuous improvement train keeps on rolling, a few new capabilities were built and also made more easily consumable by non-technical people. Recent Updates Over the last week a series of small improvements were made to the content analysis…

[mathieuisabel]

Dynamic Structuring of Content Analysis

While working on the next evolution of how content is processed on my personal project with Dretza, I had to design and build new APIs to assist in that area. These APIs offer powerful capabilities for generating assertions, determining critical aspects for content reviews and automate the content analysis process itself. There was also a few updates to the content assertion API to improve…

[mathieuisabel]

Navigating Grey Zones - Consistently Picking the Right Tasks

In our previous discussion, we explored the versatility of the Solver API in solving complex decision-making problems. Now, let’s dive deeper into a specific scenario: prioritizing tasks within an organization by considering multiple dynamic factors. By harnessing the Solver API’s power, businesses can not only streamline task allocation but also adapt in real-time to changing business…

[mathieuisabel]

A New Era of LLM-Powered Negotiations

As union and government negotiations are a hot topic here in Quebec these days, I figured I would use that as a toy example in my exploration of the Microsoft Autogen framework.

Introduction In a world increasingly captivated by the potential of AI, Microsoft’s recent framework, AutoGen, stands out. AutoGen revolutionizes how we view and interact with large language models (LLMs). It’s not just an advancement; it’s a gateway to a new dimension where AI entities, known as agents, can communicate among themselves to solve complex tasks. The real kicker? These agents are…

[mathieuisabel]

When Retrieval Augmentation Generation Doesn't Quite Cut It

In this new post, I'm discussing how a new API I've developed helps address a particular challenge faced with the Retrieval Augmentation Generation technique.

One approach to answering user inquiries with LLMs and reference material is called Retrieval Augmentation Generation, RAG for short. While RAG can serve in a multitude of question/answering scenarios, it can fall short in certain situations. A common challenge in conversation-based applications that leverages Large Language Models is constraining responses to a valid set of options that have…

[mathieuisabel]

Navigating Grey Zones In Decision-Making: Introducing the Solver API

In today’s complex and ever-changing world, making decisions is rarely straightforward. Often, the best solution is not perfect but requires careful balancing of various factors and compromises. This is where the Solver API comes into play, enabling you to score and find the optimal option from a myriad of possibilities while considering a wide range of requirements. In this blog post, we’ll…

[mathieuisabel]

Doing assertions on unstructured content

In today’s fast-paced world, dealing with unstructured text data can be a daunting task, especially when you need to extract valuable insights or validate specific information. Whether you’re managing product catalogs or verifying expense reports, ensuring data accuracy is essential. That’s where a custom-built API can come to the rescue. In this blog post, we’ll explore a novel API developed for…

[mathieuisabel]

When you need more than layout specific extraction (part 2)

As requested, a second part that goes a little more in depth as to how extraction pipeline functions.

There you go! A bit more technical details about the extraction pipeline covered in the introductory article When you need more than layout specific extraction. High Level Process Extract Text Content and Layout In order for extraction to be performed, you need to extract the text content if the source is an image. When doing so, it’s important to maintain the structural elements of the…

[mathieuisabel]

When you need more than layout specific extraction

Over the last few years, I had the opportunity to be involved in multiple projects revolving around extracting structured information out of unstructured or semi-structured content. I was able to experiment with different tools with each their own advantages and drawbacks. At the beginning of 2023, I embarked on a new personal adventure with the development of a new web site. To say it’s a…