huibert-aalbers.com
My home away from home
AI Demos
This is a collection of AI apps that my team and I use to demonstrate basic AI functionality to my customers, using local AI models with Ollama.
Simple AI demos
This macOS app includes two simple demos on how to use AI for Sentiment Analysis and Sentence Similarity. In order to use the program you need to first install Ollama and make sure that you have at least two models installed. The first one will be used for sentiment analysis. I suggest you use granite4-tiny-h. It is an open model from IBM that is small enough to perform quickly on a laptop yet it supports multiple languages. The second model is needed for the Sentence Similarity demo. It is an embedding model. I suggest you install granite-embeddings, also from IBM, that generates a 384 dimensions vector from text, which is enough for relatively short text sizes (like the ones you are likely to use for this demo).
Demos Data Uploader
The next demos require a vector database to store the data used by the apps. I have worked with several vector databases and the one I enjoy using most is SingleStore. You can get a free account and use this application to upload the data that these demos require. It will create three tables and upload the data. Uploading data for the Quora app will take several hours in order to upload the 100,000 records used by the app but the two other app's data is uploaded in seconds.
In addition to the database, all apps also use a Python script to connect to the database. You will need to install OrbStack on your Mac to run the provided container that will deliver the web services required to connect to the back-end database.
Run this command to start the container: docker run --name demos -d -e DB_USER="your_user" -e DB_PASSWORD="your_password" -e DB_HOST="svc-...svc.singlestore.com" -e DB_PORT="3333" -e DB_NAME="database_name" fastapiproject1-image:latest
Download Demos Data Uploader 1.0
Quora
This demo shows how semantic search works. It uses an open dataset published by Quora with 100,000 records stored in SingleStore table that the previous app will create automatically for you. This demo uses granite:embeddings, a free multilingual transformer from IBM to generate the embeddings locally.
RAG
RAG stands for Retrieval Augmented Generation. It was a first attempt to make LLMs more useful by allowing them to work with the user's data and is specially important in enterprise environments. This demos illustrate the concept by allowing the user to search an old 80's Apple II maintenance book. It is meant to demonstrate what customers could achive by providing their employees access to all their policy and procedure documents. This demo uses granite:embeddings, a free multilingual transformer from IBM to generate the embeddings locally and granite:.
Currícula
This app demonstrates how AI can help with productivity gains by automatically extracting relevant information from resumes to quickly build a candidate search database. It uses IBM's granite-4:small-h open LLM to extract information from resumes provided in PDF format, greatly simplifying data capture, a common enterprise problem.
