AI Decision Support System

1. Problem

People starting in AI often choose a path too early, based on hype or incomplete self-assessment. That leads to mismatched learning plans, frustration, and wasted effort.

2. Solution (high level)

This repository documents a human-in-the-loop decision-support workflow that helps narrow AI entry paths using structured qualitative signals and elimination heuristics.

Status / Disclaimer: This is a documentation-led portfolio case study, not a runnable automated system or pipeline. The Power Platform content below is a conceptual mapping, not a claim of a built production solution.

3. Workflow (step-by-step)

1. Collect structured inputs across four diagnostic modules.
2. Convert notes into qualitative signals (not scores).
3. Apply elimination logic path-by-path.
4. Flag ambiguity or conflict when signals are weak.
5. Produce a recommendation + alternatives + conditions.
6. Keep final judgment with a human reviewer.

Detailed workflow:

• docs/workflow.md
• Decision logic: docs/decision-logic.md
• Decision flow diagram: docs/architecture.mmd

4. Inputs / Outputs

Inputs

• Profile context and constraints.
• Work-style preferences.
• Frustration/ambiguity tolerance.
• Expectations and available resources.

Outputs

• Decision report with:
  • recommended path,
  • eliminated paths and rationale,
  • alternatives/conditional options,
  • explicit human review reminder.

Examples:

• examples/synthetic_input.json
• examples/quick_walkthrough.md
• examples/synthetic_output.md
• docs/example-report.md

5. Edge cases

The workflow handles low-confidence scenarios explicitly, including insufficient, conflicting, or ambiguous signals, plus iteration/re-entry.

Details:

• docs/edge-cases.md

6. KPIs (planned)

No measured KPI outcomes are claimed in this repo. KPI definitions are provided as PLANNED metrics for future implementation.

Details:

• docs/metrics.md

7. Quick demo (how to read/use the example artifacts in this repo)

1. Open synthetic input: examples/synthetic_input.json.
2. Read walkthrough: examples/quick_walkthrough.md.
3. Compare with synthetic output report: examples/synthetic_output.md.
4. Cross-check logic in docs/decision-logic.md.

All synthetic_* artifacts are explicitly synthetic examples for portfolio demonstration.

8. Power Platform mapping (conceptual implementation)

If translated to Microsoft Power Platform (conceptually):

• Power Apps: guided intake form for the four diagnostic modules.
• Power Automate: orchestrates stages (signal normalization, elimination checks, draft report assembly).
• Dataverse / SharePoint lists: stores intake records, rule snapshots, and review outcomes.
• Human approval step: reviewer confirms/edits recommendation before sharing.
• Document output: report template populated and exported for user reflection.

This mapping describes implementation intent only; it does not claim this repo contains a built Power Platform app/flow.

9. What this proves (skills demonstrated)

• Business-process thinking: clear stages, entry/exit conditions, and traceable decisions.
• Automation design readiness: workflow decomposed into steps suitable for Power Automate.
• Human-in-the-loop control: explicit review checkpoints instead of blind automation.
• Documentation quality: decision logic, architecture, examples, and edge-case handling are auditable.
• Portfolio maturity: scope boundaries and non-goals are clearly stated.

Related context:

• docs/case-study-pan-m.md
• docs/portfolio-audit.md
• roadmap.md