GenAI Architecture

Implement a Python interaction with a local or free-tier language model that uses system and user messages, supplied grounding context, and an explicit output contract.

Validate model output against a JSON schema and implement handling for invalid responses.

Implement a constrained model-mediated call to one read-only SaaS API with server-side argument validation.

Split SaaS knowledge-base documents into retrievable chunks with source and tenant metadata.

Generate embeddings and store the document chunks in a local vector index.

Retrieve relevant evidence while enforcing tenant-level metadata filters.

Generate an answer grounded in retrieved evidence with verifiable source citations.

Diagram the end-to-end request, data, and trust-boundary flows for a multi-tenant GenAI SaaS application.

Map each logical GenAI architecture component to an AWS service and a cloud-agnostic alternative.

Specify timeout, retry, fallback, and degraded-mode behavior for model, retrieval, and tool dependencies.

Define service-level objectives and an observability plan covering latency, errors, token and cost usage, retrieval failures, and safety events.

Create a threat model covering direct and indirect prompt injection, cross-tenant data leakage, sensitive-data exposure, and tool misuse.

Design a mitigation matrix that applies authorization, tenant isolation, input and output controls, and least privilege to the identified threats.

Define data-retention, auditability, human-escalation, and incident-response requirements for the GenAI SaaS feature.

Build a compact evaluation dataset containing normal requests, edge cases, prompt-injection attempts, and cross-tenant access attempts for the SaaS support copilot.

Apply simple metrics, rubrics, and pass/fail checks to measure retrieval relevance, citation-grounded answers, task success, and isolation or safety behavior.

Compare at least two candidate model or service options using a scorecard for quality, latency, cost, privacy, and operational complexity.

Integrate the previously built prompting, retrieval, validation, and tool-use components into an end-to-end local SaaS support copilot.

Assemble an interview-ready architecture pack containing system and trust-boundary diagrams, key decision records, evaluation evidence, and risk and operational controls.

Conduct a mock customer-discovery discussion and present a concise solution recommendation.

Defend the architecture's trade-offs, economics, security controls, and failure handling in a timed system-design interview.

Produce three evidence-based STAR career stories and matching résumé or LinkedIn bullets that connect prior software architecture experience to GenAI architect responsibilities.