AI-Assisted Project Charter Generation Standard Operating Procedure

Standard Operating Procedure: AI-Assisted Project Charter Generation

1. Document Control Block

• Document ID: SOP-TR-ENG-042
• Effective Date: October 24, 2023
• Version: 2.1.0
• Review Cadence: Semi-Annual
• Classification: Internal Operations / Engineering Standards

---

2. Executive Summary & Purpose

This Standard Operating Procedure (SOP) defines the institutional protocol for leveraging Large Language Models (LLMs) and automated generation pipelines to produce enterprise-grade Project Charters at Template Registry. The purpose of this protocol is to eliminate semantic drift, enforce structural consistency across all organizational initiatives, and reduce human-hour generation overhead by a minimum of 65% while maintaining absolute compliance with governance, risk, and compliance (GRC) frameworks.

---

3. Scope & Prerequisites

Scope

• Applicability: All technical, operational, and strategic initiatives originating within Template Registry.
• Boundaries: This SOP covers the ingestion of raw requirements, prompt engineering parameters, AI generation mechanics, and human-in-the-loop (HITL) validation. It excludes downstream execution tracking and resource allocation engines.

Prerequisites

• Software & Tooling:
  • Template Registry AI Workspace (Enterprise Tier, V4.2+)
  • Prompt Management System (PMS) access
  • Jira / Linear API integration token (Read/Write)
  • Confluence Enterprise Documentation Suite
• Artifacts:
  • Approved Business Case Document (BCD)
  • Stakeholder Matrix Draft
  • High-Level Requirements Document (HLRD)

---

4. Roles & Responsibilities (RACI Matrix)

---

5. Step-by-Step Procedure

Phase 1: Context Ingestion and Data Sanitization

• 1.1 Export the approved Business Case Document (BCD) and High-Level Requirements Document (HLRD) into clean Markdown (.md) format.
• 1.2 Run the Data Sanitization Script (scripts/sanitizer.py) to purge proprietary secrets, PII, and unreleased pricing structures from source files.
• 1.3 Verify that the sanitized payload conforms to the maximum token window constraint of the target LLM (hard cap: 16,384 input tokens).

Phase 2: Parameterization and Prompt Execution

• 2.1 Access the Template Registry Prompt Management System and load baseline schema PROMPT-CHART-GEN-v3.yaml.
• 2.2 Inject sanitized operational context into system prompt variables: {{PROJECT_NAME}}, {{BUDGET_CAP}}, {{TARGET_TIMELINE}}, and {{RISK_TOLERANCE}}.
• 2.3 Set generation hyper-parameters: Temperature = 0.2 (for factual deterministic adherence), Top-P = 0.9, and Frequency Penalty = 0.0.
• 2.4 Execute the generation pipeline and capture the output payload ID.

Phase 3: Human-in-the-Loop (HITL) Validation & Structural Audit

• 3.1 Perform automated schema validation against the Template Registry Charter Schema v4 (JSON Schema check).
• 3.2 Audit the generated Executive Summary for hallucinated milestones or unauthorized resource commitments.
• 3.3 Confirm that the RACI matrix generated by the AI aligns strictly with organizational authority thresholds.
• 3.4 Route the draft charter through the compliance checker to ensure regulatory metadata is fully realized.

Phase 4: Final Sign-Off and Registry Ingestion

• 4.1 Obtain asynchronous approval signatures from the Project Sponsor and the Chief Architect.
• 4.2 Commit the finalized project charter into the central Confluence repository under the designated project space.
• 4.3 Trigger the webhook to initialize downstream project workspaces in Jira and the resource allocation matrix.

---

6. Quality Assurance & Pro-Tips

Best Practices

• Deterministic Prompting: Always maintain low temperature settings (0.1 to 0.3). Creativity is not desired when establishing contractual project parameters and governance guardrails.
• Iterative Ingestion: If a project scope is excessively broad, do not attempt single-pass generation. Ingest by functional epic (e.g., Infrastructure first, then Security, then Product UI).

Common Pitfalls

• Hallucinated Dependencies: LLMs frequently invent technical dependencies that do not exist within the infrastructure catalog. Cross-reference all technical dependencies against the Template Registry Enterprise Architecture Blueprint.
• Scope Creep Injection: Ensure the AI prompt explicitly includes negative constraints (e.g., "Do not assume third-party SaaS integrations without explicit mention in BCD").

Metric Thresholds

• Generation Time: $\le 120$ seconds from ingestion to raw draft.
• HITL Correction Rate: $\le 15%$ text modification required by the human reviewer.
• Schema Compliance: $100%$ pass rate on automated JSON schema validation.

---

7. Frequently Asked Questions

Q1: What should I do if the AI-generated charter includes budget figures that contradict the Business Case Document?
A: Immediately halt the validation phase. Adjust the system prompt weight for financial constraints, or manually inject the exact budget ceiling into the system variables, and re-run Phase 2. The BCD takes absolute precedence over AI-extrapolated figures.

Q2: How do we handle edge-case projects that do not fit the standard Template Registry charter taxonomy?
A: Escalate to the Chief Architect's office. You will be provided with an auxiliary schema (PROMPT-CHART-EDGE-v1.yaml) designed for exploratory R&D initiatives, which relaxes structural constraints while maintaining core governance minimums.