1. Principles
Our objective is to produce useful, defensible, and outcome‑driven recommendations. We prioritise clarity, risk‑awareness, and measurable impact.
- Human‑in‑the‑loop for material decisions and deliverables.
- Security‑first: minimise, compartmentalise, and audit sensitive data.
- Traceability: preserve prompts, versions, and key assumptions.
- Continuous improvement: evaluate outputs and iterate against KPIs.
2. Governance
Design & oversight
- Pattern libraries and templates reviewed by VSCG.
- Model selection based on task class, cost, latency, and privacy.
- Evaluation sets for regression checks on key outputs.
Operational controls
- Prompt injection & jailbreak awareness; guardrails for tools.
- Rate‑limits, retries, and circuit breakers to ensure resilience.
- Fallback paths to human review for edge cases.
3. Data handling
- Avoid pasting secrets, credentials, or regulated identifiers.
- Where necessary, redact or pseudonymise prior to submission.
- Third‑party AI infrastructure may process inputs; you remain the data controller.
- Store only what’s needed for quality, audit, or compliance.
4. Model usage
We select models appropriate to the task (reasoning, generation, extraction). When feasible we use vendor features that minimise data retention.
- Document assumptions, sources, and uncertainty where relevant.
- Prefer deterministic pipelines for calculations and compliance outputs.
- Summaries are suggestions, not mandates; request alternatives when needed.
5. Transparency & limitations
AI outputs can be incorrect, incomplete, or outdated and are not legal, financial, medical or professional advice.
- Independently verify critical figures, compliance items, and security controls.
- Use human review wherever outcomes materially affect safety, cost, or risk.
6. Responsibility
You are responsible for decisions made using AI‑generated content. Apply appropriate review, sign‑off, and change‑control.
- Retain accountability with named owners and approvals.
- Record context and rationale for material changes.
7. Using AI effectively
Ask for the right shape
- Objectives, constraints, timelines and success metrics.
- Alternatives, trade‑offs, risks, cost estimates.
- Action plan with milestones, owners, and KPIs.
Make outputs practical
- Request summaries first, then expand details.
- Ask for checklists, acceptance criteria, and review steps.
- Call out assumptions and dependencies explicitly.
8. Contact
For questions about this guidance or to discuss enterprise AI strategy and governance, contact hello@vishaalsingh.com.