Gurov AI brain dump

[gurov]

Howdy folks. Here's a dump of the AI toolset i've used today to implement some useful-to-me features. This post isn't about features, but about the process by which they were willed into being.

I use two IDEs, Clion for when i actually have to touch code, and Cursor. I got the pro upgrade for like $200/year. I compile locally using my build container. Build for me takes 42 seconds for a full bundle, so iterations on code changes are incredibly quick in this tight dev loop.

There is one file i ended up writing by hand at first, i called it ai_hints.md, it is attached to this post. dumping some basic knowledge of the codebase/structure/constructs/concepts/terminology/purpose into it. I believe that helped improve some of the results.

The task for today, driven by my incredibly selfish and self-serving desires for happier engine management were:

1. non-linear tachometer multiplier, being able to trim what gets output to tachometer per RPM, to correct for weird old hardware, or shenanigans
2. speedometer startup sweep
3. speedometer non-linear multiplier, for the same purposes as the tacho non-linear multiplier
4. ETB feedforward TPS vs BATV correction, to trim ETB authority for dead/dying alternators/lights/electrical loads/etc.

The tasks were completed by the agent, with some help from the human. and implementation details were dumped into 2025-08-06.md.txt (attached to the post). As a bonus, it converted a legacy module to EngineModule framework, mostly correctly, and fixes to that were captured in the date-stamped file.

There were some hand fixes that had to be applied, and there were learnings relayed to the agent (these are captured in the date-stamped file, it created at my request).

At this point, I think these simple modifications can be achieved by the ai agents under supervision. It doesn't get it perfect, but it comes pretty damn close.

[Automate]

For limited development, can the free Cursor account leverage what you have done, or does it take Pro?

[mi-hol]

This post is ... about the process by which they were willed into being.

I use two IDEs, Clion for when i actually have to touch code, and Cursor.
...

There is one file i ended up writing by hand at first, i called it ai_hints.md, it is attached to this post. dumping some basic knowledge of the codebase/structure/constructs/concepts/terminology/purpose into it. I believe that helped improve some of the results.

The tasks were completed by the agent, with some help from the human. and implementation details were dumped into 2025-08-06.md.txt (

Hi gurov,
thanks for sharing your success with AI coding using your preferred IDEs.
I'm not into that topic yet, but am very much interested to learn.

Would you mind to elaborate on how "... ai_hints.md helped improve some of the results."?
Maybe via a screen cast/video of this preparation step, feeding it into the used AI?

Thanks Michael

[gurov]

For limited development, can the free Cursor account leverage what you have done, or does it take Pro?

free account will run out of quote pretty quickly, but you could get a preview

[gurov]

This post is ... about the process by which they were willed into being.

I use two IDEs, Clion for when i actually have to touch code, and Cursor.
...

There is one file i ended up writing by hand at first, i called it ai_hints.md, it is attached to this post. dumping some basic knowledge of the codebase/structure/constructs/concepts/terminology/purpose into it. I believe that helped improve some of the results.

The tasks were completed by the agent, with some help from the human. and implementation details were dumped into 2025-08-06.md.txt (

Hi gurov,
thanks for sharing your success with AI coding using your preferred IDEs.
I'm not into that topic yet, but am very much interested to learn.

Would you mind to elaborate on how "... ai_hints.md helped improve some of the results."?
Maybe via a screen cast/video of this preparation step, feeding it into the used AI?

Thanks Michael

it's a pretty simple step. both humans and machines need documentation. LLMs need documentation to fill their context window with relevant details. It needs quite a bit of input in terms of hints, so it can at least start to discover what's what.

If you want to try to replicate what i've done, load up my files i've posted into the repo, and command the model to

Code: Select all

load ai hints from ai_hints.md, from now on this is the file to reference when starting a new task. scan the repository and validate hints[code] or something like that. you should see the context window fill:
 
 [attachment=0]image.png[/attachment]
 
then you have to take it step by step and validate its output, instructing it to update context files, create new ones with learning from the sessions.