AI + Dev + Life

AI is taking over the world. I've been working on LLM-based AI app development for the past year or so. I am familiar with open-source libraries like LangChain, LangGraph, and LlamaIndex, and other more "standard" AI SDKs from OpenAI and Google, etc. Also, I'm no expert, but I have a good understanding of prompt engineering. Anybody who wants to start LLM development should first start learning effective prompting.

The most common LLM applications are custom chatbots augmented with domain-specific knowledge, e.g., primarily through RAG (retrieval augmented generation). RAG is more of an art than a science. There are a lot of factors to consider to build an effective RAG system such as chunking, choice of embeddings and vector databases, etc. And, ultimately, understanding and managing "context", and not just prompts, is required to build effective chatbots.

A new kind of LLM-based apps, generally called 'AI Agent', are getting a lot of attention these days. They are typically characterized by the fact that they use "tools", e.g., relying on MCP (model context protocol), often based on the decision made by the agents themselves. Hence we sometimes call them "autonomous AI agents", but this characterization is debatable. "Deep research agent" is one of the fastest developing areas.

Besides chatbots and autonomous AI agents, I've done some general LLM development, such as quantization and fine-tuning, etc., in addition to RAG. Also, LLM-based document analysis is really an application of RAG. What else can we do with LLM technologies?

I've been doing fullstack web app development for many years now. I started as a backend developer, and picked up frontend development over the years. For backend development, I mostly use Java, C#/.Net, Go, Python, and Javascript/Typescript (Node.js). For frontend, I generally use frameworks like Angular and React/Next.js. I also use various UI libraries such as Tailwind CSS and Shadcn UI. Web development is really my bread and butter. Despite the fact that I may be rusty in some of the technologies, this is the areas I can deliver consistently, whatever the requirements are.

I've been using various services of AWS and Google Cloud for many years now, and to a lesser degree other cloud platforms like Azure. (The first Google Cloud service I used was App Engine (GAE), back in 2011. At the time, I was working on a URL shortener service, and the cold start problem of GAE was a big issue. (Oh my my. How many different tricks I tried to beat the cold start problem?) On top of this, Guido van Rossum refused to support naked domains. The project was ultimately abandoned. It's funny that I remember these things after all these years. 😄)

I've used, quite extensively, pretty much all widely-used databases like Postgres and MongoDB, and Redis, etc. I am also familiar with various cloud-based database services such as AWS RDS, DynamoDB, Aurora, and Google Cloud Firestore and CloudSQL. I have been also using vector databases like Chroma and Pinecone recently as well as the pgvector extension of Postgres.

I did Android app development for many years, until a few years ago. (Haven't done it much recently. Last time I worked on Android apps was 2018. There is some story as to why I stopped doing Android development at least for my own projects, which I will not go into here. On a related note, I also stopped selling my books on Amazon bookstore. We are living in a twisted world where bullies rule the world, at every level of society, because very few people stand up to them. I believe that somebody has to.) I also have some limited experience with cross-platform tools like React Native and Flutter.