Carl Free
AI Engineer / Machine Learning Engineer
Click here for resume'.

 

A Few Examples of His Work
(with downloads and video demos!)
Ai Enterprise Software Generation This is not vibe coding. Carl created an Ai factory that translates human-written specification documents into finished and fully tested software. It is designed for medium to large-scale projects -- from tens of thousands to hundreds of thousands of lines of code. The output is reliable, repeatable, and easier to maintain than hand-written software.

Demonstration video here
Virtual Employee Carl created a software product that runs on a VM and can control not only browsers, but desktop software, control panels, and anything else a human user can control. The Virtual Employee accomplishes this by operating a Remote Desktop (RDP) session, giving it the ability to interact with any application just as a person would.

See a presentation about this product (video demos are embedded) here.
Specialized RAG System with Heirarchy Carl recently created a retrieval augmented generation (aka "RAG") system which has a unique feature: it instantly understands a heirarchy of departments, projects, ventures, and experiments. This is essentially an Ai-enhanced repository for an entire enterprise, which allows each user to access and update only his/her sub-organization's documents. See step #5 below to get the unique feature of this RAG system.
  1. When user logs-in, they see their own department's documents.
  2. Users can upload documents into their own department's library, which are used to inform an Ai.
  3. Users can chat with an Ai (i.e. and LLM) which also has access to their department's library.
    This allows the user to ask questions like, "Which interviews require a follow up meeting?", or "What requests for help have not been properly answered?"
  4. Users can create a sub-organization around a department, project, or anything else.
  5. Now here is the unique feature: The LLM from step #3 can also see all the sub-organizations' documents.
    This allows managers to ask questions like, "What are the worst pain points my organization is experiencing?" or even, in the same chat, "What are the new pain points since the last time I asked?" Making the system aware of the heirarchy allows chat with any combination of documents uploaded to the user's organization and all organizations below. This means a member of the root organization can ask questions about the whole enterprise, which are answered intelligently with up-to-the-minute information.
Carl originally wrote his own vector database from scratch to inform the LLM, but when better vector databases came to exist, he swapped this out for one of them.
Intelligent Bot that Operates Apps Carl recently created a bot that takes human-language commands from a messaging app, and uses various online and locally-installed apps to execute those commands.

Demonstration video, here.
Reconstructing Legacy Software with AI Carl recently trained a Neural Network with PyTorch, that examines an old legacy application, learns its behavior, and then reproduces that behavior as a modern web app!

Demonstration video, here.
Loading a Neural Network into a Smartphone App Carl wanted to see if he could get a cellphone to do inference without any connection to a server, so he trained a suitable Tensorflow model with the classic MNIST dataset, embedded it into a Flutter app, and deployed it to Android OS. Note: Unlike the rest of the works in this showcase, which are completely original, Carl followed an online course to do this.

A video of the app being used, is here.
Download the app itself here.
Video Enhancement with AI Carl assembled some very good AI libraries and a pre-trained image classifier, and used them to annotate videos.

Watch a very short sample of the result (including a selfie) here.
Data Science Bot To make data science products more accessible, Carl created a bot that can do the setup and configuration of IBM's Cloud Pak for Data accounts.

See a small example of what this bot can do: here.
Earth Monitor Carl wrote this beautiful application for a company that monitors power-generation systems all over the world. The company needed something that humans could tolerate seeing for many relatively uninterupted hours. It was displayed on a large monitor in the middle of a control room, and on smaller monitors in the bathrooms and break rooms, so there would be eyes on it nonstop.

The application took a flat map of the earth, plus many points of lattitude and longitude and status, and wrapped them around a rotating sphere. At the time, there was no viable 3D or OpenGL option for Java, and the company wanted the option to switch platforms, so Carl had to use calculus and 3D trigonomitry, and had to optimize those calculations to be done quickly in real time.

The speed, direction, and tilt of the rotation could be controlled with arrow keys. The alarms and warnings had sounds, which could be silenced by clicking on the text or by clicking on the blinking dots on the rotating earth.

Download a video of this program at work here.
Grade Maker Also quite beautiful, this shrink-wrapped PC application was sold online and from TV commercials. It helped children achieve more in school. Carl did all of the design and programming, but had the help of a talented artist for the graphics.

Download a video of this program at work here.
Download the installer here.
COBOL to SQL translator The client had many types of data stored in a multi-million-line COBOL application, and needed all of it migrated to a relational database. Since the result would be more than 2000 database tables, doing this translation manually would have required a huge number of man hours and would surely have had many errors. The client had the COBOL source code, so Carl wrote a program that read the COBOL to learn the structure of the data. The program used the structure information to create an equivalent relational schema, and used the structure information again to read the raw COBOL data files, and inserted the data into the new tables.

Once written, this program did in a few hours what would have taken a typical programmer weeks or months.
Pluggable Platform The company was suffering from maintaining a number of customized versions of its LCMS software. Every time a bug was fixed in the main code line, that fix had to be re-applied to the code of every custom version. Even worse, whenever the main code line was upgraded, engineers had to start with the upgraded code line and then re-apply customizations in order to achieve upgraded custom versions.

Carl was tasked with creating a new technology for eliminating or greatly reducing this rework.

He figured out how to make customizations in the form of plugins. These plugins were not only additional components, but they could "infect" the rest of the software just like a virus would, modifying nearly any behavior of the core software, but without changing *any* of the core code. At upgrade time or bug-fixing time, simply remove the plugins, upgrade or patch the core software, then re-apply the plugins. Most of the time, no rework is needed.

This platform is used today, and it has saved an enormous number of engineering hours.
Crypto Bot Thanks to a fascination with new technologies in the area of finance, Carl has written several crypto-currency trading bots. For quick development time, Carl used Python and a litle bit of GoLang.

This didn't change the world, and didn't even change his own life much, but Carl enjoyed the learning experience.

 

Personal Information
Carl worked his way through college by teaching Ballroom dance lessons during evenings and weekends. He still teaches and dances in his spare time. Carl grew up in Central Florida but lives in Ireland now. Carl enjoys dancing, teaching, and public speaking.

Carl got his Bachelor's degree in Computer Science at the University of Central Florida, and started his post-graduate studies with emphasis in Artificial Intelligence, but got a bit disillusioned with AI, and never completed his Master's degree.