How your data is handled — in plain English.

• Your data, your control. You own everything I build and every dataset I train on for you.
• No selling, no sharing. Client data is never used to train shared models, sold, or shared with third parties beyond the operational services listed below.
• Cost and limits documented up front. What each provider costs monthly, and what it will and won't do, is written down before we start.
• You can leave. Everything runs on standard, portable tools. When you want to switch providers or take the whole thing in-house, the code and data come with you.

For the systems I typically build, here's where data actually travels. This is the default stack — many parts can be swapped based on your requirements.

• Data you don't share with me — I only receive what's explicitly sent for a specific system.
• Credentials, API keys, and secrets — always stored in your infrastructure (Vercel, AWS Secrets Manager, or your equivalent), never in my working copies.
• Source data for internal tools — by default, custom internal systems query your existing databases directly. No copies land in my environment.

• Audit form answers: kept 12 months, then anonymized. You can request deletion any time by emailing me.
• Chatbot conversations: retained per the policy we set together (default: 30 days). Users can request their conversations be deleted at any time.
• Project data: on final delivery, all code and configuration transfer to your team. Any working copies on my machines are wiped within 30 days.

For clients with regulatory requirements (healthcare, finance) or a strong preference for control, I can deploy on your own infrastructure. That typically means: Postgres on your servers instead of Supabase Cloud, your own Cloudflare or AWS account instead of Vercel, and Azure OpenAI (with private endpoints) or self-hosted open-source models instead of the shared APIs. This adds engineering time and monthly infrastructure cost — worth it for the right situation, not for most small businesses. I'll be honest about which category you're in.

This is the most important section. If any of these apply, I'll say so during the audit — a lot of AI projects should never have shipped.

• Legal, medical, or safety-critical judgments where a wrong answer causes real harm and no human is in the loop to catch it.
• Compliance-sensitive workflows where the audit trail and explainability requirements exceed what current models can defensibly meet.
• Volumes so low that automation costs more per month than the human hours it replaces (this is more common than people think).
• Problems that look like AI problems but are actually workflow, staffing, or process problems in disguise. Adding an agent to a broken process makes it break faster.
• Anything the customer or user needs to feel a human is on the other end. Some interactions should not be automated even when they can be.

I'm an AI systems specialist. I'm not a lawyer, a tax advisor, an HR compliance consultant, or a security auditor. I can help you meet data-handling requirements you've already defined, but I can't tell you what those requirements are — you'll want a qualified professional in your jurisdiction for that (PIPEDA in Canada, GDPR in Europe, HIPAA in the US, and so on). I'll flag things that look like risks, but the responsibility for compliance sits with you.

If any of this needs clarifying for your specific situation, email me before you engage me. It's faster than working it out in the middle of a project.