Research papers and publications on AI-assisted software development methodology, architecture governance, and verification.
Two connected papers on AI-assisted software development methodology, and an open-source tool from the practice's research into small, domain-expert models.
AI shifts the binding constraint from implementation to specification. Neither Waterfall nor Agile ports cleanly into the AI era. The paper identifies the hybrid pattern best-performing teams are converging on: Agile cadence, Waterfall-quality specification, AI implementation.
Download PDFIntroduces the SBR framework and its central concept: the trust boundary — the provability boundary of the test suite, defining the region within which autonomous AI generation is verifiable. The framework integrates EARS requirements, spec-driven methodology, architecture governance, TDD, and BDD into a directed chain for AI-assisted development.
Download PDFBuilding a small LLM with domain-specific knowledge and enterprise IP, embedded in a high-performance application, is the next frontier of AI applications — and curating good fine-tuning data is the biggest hurdle. tkn_perplexity is a visualization tool that helps AI developers see the difference between good and bad training data.
Get Access on GitHubThe papers above have been reviewed by colleagues and are published here. The tkn_perplexity repository is open source on GitHub.
Marigold Intelligence LLC
inquiry@marigoldintelligence.us