Open Research.
We believe in building in public. This means sharing our successes, but more importantly, documenting our failures. Below are the experimental projects we've undertaken and the hard lessons learned.
Thinking Patterns
Pivot / Failed ExperimentOur 1st attempt to fully automate the logic structuring of strategic AI reasoning frameworks using Ollama, Claude Code, and Qwen 3.6.
The Hypothesis
We theorized that by feeding complex cognitive schemas into automated code-generation loops, we could rapidly produce robust SDKs (`llm-thinking-patterns-nakprc`, `nakprc-thinking-patterns`) without manual intervention.
The Failure
The 1st attempt failed in just 1 day. Current AI models (Qwen 3.6 via Claude Code and Ollama) failed to consistently manage the depth of logic required. The automated output yielded inconsistent, hallucinated, and non-deterministic logic structures, breaking the core reliability needed for an SDK.
The Pivot: Attempt #2
We are actively learning how LLMs operate at a deeper level. The project is not abandoned—we will try a 2nd attempt using a hybrid approach combining automated AI code generation with rigorous manual engineering to ensure absolute fidelity in our packages.