expert-system-engineer

Expert System Engineering

Full lifecycle methodology for engineering knowledge-based expert systems. Upstream phases use CommonKADS for organizational scoping and conceptual modeling. Downstream phases use generalized MYCIN/EMYCIN patterns for rule-based implementation. Follow all phases in order.

For domain terminology, see terminology.md.

Phase 0: Suitability Assessment

Determine whether the problem warrants an expert system before committing to the methodology.

1. Classify the problem type:

   • Analytic (classification, diagnosis, assessment, monitoring) — system interprets existing data
   • Synthetic (configuration, planning, scheduling, design) — system constructs a solution from components
   • Neither — redirect to a different approach

2. Evaluate expert-system fit against alternatives:

   Criterion	Expert System Fit	Alternative
   Domain has identifiable human experts	Required	If no experts → data-driven ML
   Knowledge is heuristic, judgmental	Strong fit	If purely algorithmic → conventional software
   Reasoning must be explainable	Strong fit	If black-box acceptable → ML/statistical
   Domain is bounded and well-scoped	Required	If open-ended → LLM-based approaches
   Decisions involve uncertainty	Strong fit (CF/rules)	If precise probabilities needed → Bayesian networks
   Training data is limited	Advantage over ML	If large labeled datasets exist → ML

3. Make an explicit go/no-go decision. Document the rationale.

Completion gate

• Problem type classified (analytic or synthetic)
• Fit evaluated against at least three alternatives
• Go/no-go decision documented with rationale

Phase 1: Organizational Scoping & Feasibility

Scrutinize the organizational environment, isolate the knowledge bottleneck, and confirm feasibility before technical work begins.

1. Map organizational context (OM-1): Document business mission, drivers, problems, opportunities, and candidate solution directions.