posts

In Graph databases, data is stored as nodes and relationships instead of tables, forming a graph where nodes (vertices) connect via relationships (edges) in a labeled property graph. Category of these nodes are called Labels.

Before exploring Project Babylon’s details, we will look into some of the fundamentals of the GPU programming model, using NVIDIA’s CUDA as reference and whenever GPPU mentionedin this article, it implies NVIDIA’s CUDA-enabled GPU.

This is a follow up of my previous article. Here, I demonstrate how we could set up multiple MCP client-servers and how an agentic pattern can be used to integrate them.

This document briefly describe how a MCP server that can connect to the Perplexity chat model can be created using Spring boot and how to use mcp inspector as MCP client to interact with them.

This document birefly describe how a RAG setup is achived using PgVector, Ollama and Perplexity cloud APIs. PgVector is an extension of PostgresSql data base where we can store embedding of our data which we can be used to provide context for a RAG. Unlike Gemini, Peplexity does not have an embeding model (atleast for now) that we can use. Therefore, we can use a locally running Ollama to create the vector embdedding of any given text. Once, we have a vector embedding for our data, we can use that as the context for any given query and pass along to the Perplexity cloud API.

Follow this guide from spring and create a sample application. I have also added spring-boot-starter-web and spring-ai-starter-model-vertex-ai-gemini etc. as the dependency. Then, download the project and open it in your favorite IDE. I have chosen gradle instead of maven. The dependency block in the build.gradle will look like something below.

Disclaimer: Do not use this write up as a 101 source for virtual threads but instead read it once you have read other 101 write ups about it to clarify some of your lingering questions in your mind afterwards.

I have recently read the api vs implementaion configuration of gradle here. Then I thought that it will be very useful to create some simple java programs which can easily explain the difference between compile time depencies and run time depencies. For that purpose, I have created some dummy classes just to intoroduce the concepts.

When it comes to programming language, Java is my goto langauge. But when I would like to try some sample realed to NLP or machine learning, I am compelled to use python because most of the examples were in python. As I follow the examples, I will try to do pip install as said in the examples. When I started doing different categories of examples (spacy, rasa, scikit etc..), I found that different libaries were depending on different version of libary and I messed up all my python installation with different version of those libary. Then I realized that it’s easier to use different python environment for different type of problems. Then I came to read a wonderful article about ‘pyenv’ and started using that. As I used pyenev further, I realized that eventhough I screwup that particular version of python it does not screw up all the system and I can live with other version peacefully. I could not find the original reference which I read that time hence here I can not point to that. Here I will be some other refer some other source which I found just before installing pyenv in a new system [1]. All the following commands are for mac os.

I have recently developed a custom rasa chat bot by following rasa tutorial. I went bit further than explained in the tutorial and created custom action which depend on a duckling server and connected a chat ui to talk to the bot. When I tried to deploy the all these together using docker-compose, I could not find a resource which directly explain how to create custom images of my chat bot and deploy it. Therefore I thought of sharing docker files which I used. I am not going through how to create a custom chat bot because that’s nicely explained here.

I have recently come across a rather cool feature of gson. I like to call it as selective deserialization. It’s easier if I explain it with the help of an example. Please see the json of a Person object with age, name, gender and address fields.

I have recently developed a demo springboot (version 2.2.5) application which use java 11 (using modular system introduced in java 9), an elasticsearch (version and 6.6.1) and mysql (version 8.0.17). Then I went ahead and tried to deploy it in a local kubernetes cluster (using minikube). Although, there were numerous tutorials about deployment in kubernetes, I could not find a sample application which cover all the aspects which I requires in a single place. In this article, I will briefly go through the some of the issues which I faced during the deployment and the solutions for that ( ofcourse, which I got it from other sources). I am not going through all the details of the kubernetes deployment file which I used because you can find the details in so many other places.