Stark from Earth @ LS @tonyfromls@qoto.org
Was it yesterday or the other day when I met my physics friend from Germany at some metaverse.
This lady was my friend and I helped her establish the needed mathematical proofs for her ideas on some graduate level physics theories.
We had a debate from last year about limits but basically because she was coming from a Physics background and not Mathematics, she was tempted to break some theorems in limits. In the end I was able to push her to publish about it and she did with her husband's help I think or some other fella.
So last time we had a chit chat about life and then on AI and she was surprised that I have made my steps towards AI Engineering but I was mostly impressed because she's already ahead in AI Engineering and from that discussion I got curious about this LLM platform she is using which seems to be fast. It is llama.cpp.
Let see if I can get some something out of using llama.cpp in hosting local AI models as my coding assistant for VS Code.
How people see higher educational degrees: You must be intelligent to be on it!
How I see higher educational degrees: I'm not intelligent, hence I need one.
🤣
Funny that while I have tested quantized version of qwen2.5-coder:14b and the 7b, the former is breaking, the latter does not know what to do with my simple instruction.
Then i noticed a little chat box on the lower right and for a while I forgot that VS Code comes with Github Copilot and I gave my instruction which was executed in split second perfectly.
Frontier models are still better than local open source ones or maybe I have not explored around yet but simply because hardware is the constraint for me at the moment.
Need to invest on a hardware infrastructure for AI for my engineering needs.
Deferred the idea on VS Code + AirLLM for now and instead, I am currently exploring other ways to integrate local AI models into VS Code.
Working on integrating VS Code with AirLLM.
Jarvis is not welcoming the idea that I go VS Code + AirLLM route but I want to so let see 😅
We are back on researching the best IDE integration with AI Coding Agent so I don't need to pay for tokens 😆
I remember that project from the bank where I was working for last year. A distributed services for reconciliation of financial data and was composed of 3 microservices and each microservice involves 2 engineers. The project involved 6 engineers. We were engineers that time, software solutions engineers.
We finished that in 6 months time and that is 12 sprints (1 sprint in our organization is 2 weeks).
Now, if I will redo that project alone and me being an AI Engineer and not a software solutions engineer anymore, I can finish that project in 3 to 4 sprints only and alone 😆.
Also there was this Indian guy who was a deep specialist in the company's proprietary technology stack and he has to maintain power of his position and value in the company so he navigated the office politics in a way that Filipino engineers will always be under his feet. Such is made possible by deep specialty in humans. His only weakness was logic so he makes mistakes on analysis and that gap I filled up so the team can deliver and get thru walls.
AI has made deep specialization of humans obsolete. The same reason why I realized I had to ditch the traditional BSEE degree studies here in PH 😆.
So, I wanted to rest this week but coding exercise came and we can't just deliver a bare minimum or an MVP. My philosophy is to deliver production grade software solutions. Hence, I put great effort on every detail and have learned to develop software not as a software engineer anymore, but as an AI Engineer.
After I released tag v1.0.0 this dawn and submitted, I was planning to get a rest moving forward until the whole upcoming week.
Yeah, right. Only in my dreams and can't rest because now I need to understand some stuffs on cutting edge AI models that can be run locally, continue my AI Engineering course, and also refresh again my technical software engineering skills.
We've become a monster in our field now that I have become a literally, unrecognizable compared to me back in 2025 as a software engineer.
Now, I have become a monster because of AI Engineering. I just hope I don't become an AI Vampire 😆
Checking out cutting edge AI developments at the moment.
One interest I want to understand is AI models with hundreds of billions of parameters that can be run on local machine.
Also got interested in AI chips that can run large parameters like the one from AMD.
v1.0.0 released!
Time to rest 💪💻🖲💽😴💤
Almost done. Just making the README but it's trivially hard 😅
H2 jdbc url changes fixed that issue on hibernate interfering with Flyway.
Added profiles for dev, test, and prod.
Now optimizing dockerization with maven layer caching.
Debugging a config smell. Hibernate is trying to make the tables while Flyway has already applied the migration files 😅
Swagger is done and looks good.
Preparing the README.md but I realized I forgot integration tests 😅.
We ain't sleeping yet. We need the integration tests.
Whew! UAT regression is done. Everything looks fine 👀
Time to add Swagger.
That was a scary change because it is significant change done in a short amount of time 😅. A lot did broke but was able to fix it.
UAT mode again, just to make sure we regression test what we did earlier in the UAT scenarios.
ONe big change coming up because apart from handling call failures on downstream systems, I also realized that all calls to downstreams are blocking 👀
Need to make it non-blocking.
Done with UAT, but then moving past scenario 10, I realized that I have not implemented failure handling on downstream system calls 😆
Working on this so I can wrap this coding exercise with the last bit on Swagger.
The best part of Dockerization is that you don't need to run every system's dependency one by one and instead they are brought to life in one go via docker containers.
