AI-Powered SaaS App Children's Home Full-Stack Developer Needed

Hi Gary,

Thanks for the invite.

Before we prepare a detailed implementation plan, we'd appreciate a little more information:

1. Have you already prepared the document structure, or should we also help define how documents are organised and indexed?

2. Which document formats will the system support initially (PDF, Word, Excel, etc.)?

3. Will organisations upload documents manually, or will there be integrations with SharePoint or other storage providers in the future?

4. Have you already decided on the hosting environment (AWS, Azure, Supabase, etc.)?

5. Is the prototype expected to include vector search and a Retrieval-Augmented Generation (RAG) pipeline, or is the document retrieval approach already defined in your technical brief?

I noticed the timeline is 6–8 weeks, which is achievable for an MVP. Do you also have a budget range in mind?

Looking forward to your reply.

1: Since this is a compliance platform for children's homes, what is your strategy for ensuring that AI responses remain legally defensible when organisational policies conflict with Ofsted guidance or statutory regulations? Should the assistant prioritise internal policy, external regulation, or explicitly surface conflicts for staff review?

2: How do you intend to prevent cross-tenant data leakage at every layer of the application and not just the database, but document indexing, vector storage, prompt construction and AI response generation? Multi-tenant AI systems often fail at those boundaries rather than in the application layer itself.

3: What happens when a policy changes after staff have relied on an earlier AI response? Should every answer remain permanently linked to the exact document version and chunk that generated it for safeguarding, compliance and audit purposes?

4: How do you want the platform to behave when the AI cannot find sufficient evidence in the uploaded policies? In a safeguarding environment, would you rather the assistant refuse to answer, escalate to management, or generate its best response with confidence scoring?

5: Will organisations eventually require completely isolated AI environments including separate vector indexes, storage and model configurations or is logical tenant isolation sufficient for your long-term commercial roadmap?

6: Are you designing this as an AI chat application, or as a regulated decision-support system? That distinction fundamentally changes how retrieval, citations, audit logging, user permissions and human approval should be architected from day one.

7: What would you consider a failed prototype? Is it inaccurate answers, poor retrieval quality, slow document indexing, tenant isolation concerns, or lack of user trust? Knowing that upfront helps prioritise the architecture around the real business risk rather than just delivering features.