Encounter of AI

 

My first experience with AI related topic was back in 2024, at the time when GenAI was a hot and hype topic. Basically GenAI (Generative Artificial Intelligence) is a type of AI that was designed to create content (text, images, video, audio or even codes) based on the input of user prompt that goes into training data (LLM) and coming back as a response to the user as newly generated context. The syntax of the user prompt is based on specially craved frameworks and there a few out there for user to comply to. 

 

After the GenAI familiarization, still fascinated by this whole AI hype, in 2025-2026 timeframe I went deeper to satisfy my curiosities and getting my hands dirty and picked up more on LLM engineering subject, nowadays most known as AI Engineering 

 

AI Engineering is a practice to build product by connecting pre-trained frontier models with APIs, databases and interface, and orchestrating with prompts, context and workflow. It involves integrating large language models, building pipelines, creating specific work components i.e. AI agents to perform multi-steps logic. These engineers are not using frontier models on their web interfaces but rather go deeper into the underlying codes to integrate and stitch the tailor-made building blocks to make them more sophisticated to solve more complex user and business requirements. 

 

Below are some mini project works I did to demonstrate I understand this topic. Python are used as the base language due to its simplicity, and its extensive libraries and frameworks make it more effective and scalable. Python codes are mostly fine to run on Jupyter notebook, but other times I need to move the codes to CoLab in order to run them due to the limitation on the CPU on my laptop (depends on the nature of the work), and I can leverage the T4 CPU box in CoLab instead. Poe API is used here as a single unified interface/API gatetway to prototype and can query multiple different frontier models without managing separate billing subscriptions. Due to the geographic location with limited access to some selected frontier models, Poe API provides a better alternative in these exercises.

​​

Comments in September 2026 to cover AI trends that impact what I did previously

​

AI technologies and evolvements change so quickly and in less than a year, the evolution of AI has changed the landscape dramatically.

 

GenAI is no longer viewed as a standalone topic and has matured and integrated into a foundational infrastructure layer powering the advanced AI systems. As frontier models are getting smarter and hence rigid acronym-based prompt formats are now less important. 

 

On the AI Engineering side, things also have changed dramatically over the past year. The introduction of MCP (Model Context Protocol) has revolutionized the way we connect; MCP is an open-source standard protocol that securely connects LLMs to external data sources and tools for better integration. We are shifting towards a more sophisticated use of RAG called Advanced RAG patterning featuring dynamic query routing which routes to different LLMs dynamically based on the complexity of the queries. This LLM routing methodology will optimize token resources and cost, complex LLM models answering complex questions while leveraging lightweight LLM models to handle the easier tasks. Finally, there is a fundamental transition from prompt engineering to context engineering. Instead of relying on static, single prompts, we’ll now leverage the entire information environments including system prompt, conversation history, vector chunks, memory caches, tool outputs as well as live data to support and ground the model’s responses. 

​

Final Comments

​

Though still amazed by what AI can do, I am not using AI daily as it might surprise a lot of people. I only use this a “tool” to assist me to a certain extent as I would like to think more deeply myself, preserving my critical thinking rather than over reliance and hence using AI as an assistant or a tool in projects instead. 

 

Recently I have spent time and explored more on AI on a personal basis to play with the vibe-coding and fast prototyping works. It is amazing and fascinates me but there are some areas we need to be cautious when doing vibe coding e.g. AI with missing business context will not know how the enterprise architecture or security would look and human involvements are needed. Coders accept code without understanding the "black-box code based" could create future issues especially in the production environment.

​

I am still thrilled by this AI topic, and have been exploring and playing with the AI related material to see what it can offer and list down all the findings for future references. I believe with good system design, sensible use, challenge their responses and responsible AI we can leverage AI as a tool to tackle some of the enterprise problems and can seamlessly optimizing workflows and improve efficiencies while digitalizing and transforming products and roadmaps.

​