Janet Xiu ShiAI Quality Evaluation Advisor | AI Educator
AI Advisory for Trusted Adoption
Quality and Governance
Powered by AWS
About
I'm Janet Xiu Shi, an AI Quality Evaluation Advisor and AI Educator based in Ireland, working with organisations internationally. Most of my work happens off this page: I embed into a business, diagnose what's actually broken, and propose the right technology, AI or not. That judgment isn't theoretical: it comes from twenty-plus years across SaaS, supply chain, FinTech, and EdTech, including 7.5 years inside Workday's own engineering quality bar.
This site is the proof layer behind that advice. The build stories below are real AWS architecture I've implemented, choosing the right services for each business rather than defaulting to one, not case studies written after the fact. Data foundations, governance, and cost discipline are decisions built into a system from day one, not checks added at the end.
Janet's Framework
A systematic approach to building trustworthy AWS and AI systems.
From discovery through delivery, each phase builds intentional architecture and governance.
Discover
Understanding requirements, data landscape, and strategic context. Deep discovery of systems, stakeholders, and constraints.
Design
Architecture design, data modeling, system patterns. Creating blueprints for scalable, governed systems.
Develop
Implementation on AWS, building pipelines, ML systems, and governance controls. Turning design into working code.
Deploy
Launching systems to production with confidence. Testing, security validation, and deployment automation.
Deliver
Measuring outcomes, optimizing performance, and ensuring trusted AI operation. Continuous improvement and governance.
Agent Boutique by Janet™
A team of specialized digital agents designed to extend strategic thinking across research, web building, quality, and marketing.
Janet Xiu Shi
AGENT BOUTIQUE
Maya by Janet Xiu Shi™
AI Advisor & AI Educator
Ming the Quality Expert
AI Evaluation & Governance
Maggie the Marketer
Marketing, Brand & PR
Selected AWS and AI Build Stories
Curated examples showing how I combine AWS architecture, data discipline, and AI governance in practice.
Architected Maya V2.0's production chat pipeline on Amazon Bedrock: an EU-region inference profile (Claude Haiku 4.5) as primary, with a cross-vendor fallback (Mistral) for provider-level outages. IAM policies scope bedrock:InvokeModel to specific inference-profile and foundation-model ARNs, not wildcard access.
EU-region inference profile for Bedrock model calls, with the application stack itself deployed in eu-west-1 (Ireland). Dedicated CloudWatch alarms distinguish Bedrock provider failures from application errors, so a model outage never silently surfaces as a generic 500.
Built Maya V2.0's knowledge base loading to keep proprietary content out of the deployment pipeline entirely: the Lambda bundle carries zero knowledge-base documents. Content is fetched from S3 at runtime, out-of-band from git and CI.
The deployment pipeline and the knowledge content are two separate concerns - one lives in code, the other never does.
Conversation continuity is stored in DynamoDB with native TTL, not an application-level cleanup job: the session record expires automatically after 24 hours, a database guarantee rather than a cron job that can silently fail. Consent receipts and saved conversation logs use the same TTL mechanism on a separate 90-day clock, matching Maya's own privacy notice. Consent receipts write to DynamoDB first, with an S3 mirror as a resilient backup.
If a database write and its backup disagree, the system trusts one of them by design, not by accident.
The static site runs behind CloudFront for global delivery speed. The API (chat, consent, health checks) deliberately does not: it stays on a regional API Gateway, close to its EU-only data and processing, rather than routed through a global edge network.
Not every endpoint benefits from a CDN. Knowing which ones should never be cached is as much a design decision as picking the services in the first place.
How I Think
Four pillars of trusted AI: strong data foundations, intelligent systems, responsible governance, and AWS credibility.
Quality data structure, observability, and readiness matter before ML or AI. Strong foundations prevent cascade failures.
Models are part of a wider lifecycle. I think in pipelines, evaluation, and feedback loops, not isolated predictions.
Responsible AI, security thinking, and practical safeguards. Built-in accountability, not retrofitted compliance.
Deep literacy across services, security, scale, and delivery trade-offs. Credible design and implementation on the platform.
From Data Foundations to AI Systems
A curated dashboard-style showcase of how structured data, ML readiness, governance checks, and cloud delivery shape intelligent systems.
Quality improves systematically when tracked and addressed.
Balanced performance across accuracy, precision, and recall.
AI Quality Governance for Trusted Adoption
High-level governance covering AI quality, assurance, guardrails, accountability, and trusted deployment.
Consent Before Collection
Analytics and any optional data collection sit behind an explicit consent action, not a default-on toggle. Nothing is collected before someone agrees to it.
Scoped by Design
Bedrock permissions are scoped to named model and inference-profile ARNs, not a wildcard grant. A system can only invoke what it was explicitly given, nothing more.
Alarms That Can't Ship Silent
The deployment pipeline refuses to provision a production alerting stack unless a real recipient is confirmed to receive it. A monitoring gate that nobody can fail is not a gate.
Failure Has an Owner
Provider-level failures, like a Bedrock outage, are distinguished from application bugs at the monitoring layer, so triage starts pointed at the right system instead of a generic error.
These are operating practices already running in production, not policy statements.
Governance that only exists on a slide doesn't count.
Let’s talk
Interested in collaboration, consultation, or learning more about how I approach AWS and AI? I’d love to connect.
© 2026 Janet Xiu Shi | AI Quality Evaluation Advisor | AI Educator. All rights reserved.