End-to-End Hands-On Project
From problem interviews to post-launch review, complete a demonstrable, reviewable project.
Reuse and Evolution / Collaboration and Product Positioning Overall Framework ↗
Table of Contents for This Page
Project Goals and Scope
Complete the project design and validation package for an enterprise after-sales ticket assistant. All enterprises, users, tickets, manuals, and metrics are synthetic materials. This case is for learning purposes only; it is not a delivered customer case, and it does not claim to run a real large model or production RAG.
The first phase supports reading tickets, searching authorized materials, generating drafts with citations, and human review; it excludes automatic sending, device control, cross-tenant queries, and automatic closure.
Six Milestones and Deliverables
| Phase | Practice Task | Review Evidence |
|---|---|---|
| Discovery | Simulated interviews and process walkthrough | Problem definition, scope, baseline plan |
| Design | Data model, components, and interfaces | Architecture diagram, permission matrix, ADRs |
| Implementation | Deterministic search first, then model integration | Reproducible run instructions and configuration |
| Validation | Normal, no-answer, unauthorized, and failure cases | Fixed samples and item-by-item reports |
| Launch Preparation | Canary rollout, alerting, and recovery drills | Release checklist and handover draft |
| Review | Value hypotheses, limitations, and reuse | Decisions, failure samples, and follow-up plans |
Downloadable Exercise Starting Point
Save the three files to the same directory and run the command below. The results of this demo only prove that the listed keywords and filter rules behave as expected on the seed samples; they do not prove semantic recall, generation quality, or end-to-end safety of a real system.
python3 offline-retrieval.py
Review and Scoring
| Dimension | Score | Evidence Required to Pass |
|---|---|---|
| Problem and Value | 20 | Facts/hypotheses separated, metrics have denominators |
| Engineering and Integration | 25 | Contracts, duplicate requests, exception behavior |
| AI and Evals | 20 | Sample versions, error taxonomy, and human scoring |
| Security and Reliability | 25 | Isolation, revocation, degradation, and recovery |
| Handover and Review | 10 | Reproduction by others and follow-up ownership |
This is a learning scoring suggestion, not a professional certification. If safety-critical items fail, the project cannot be claimed production-ready even with a high total score. Unimplemented capabilities are explicitly marked as designed or pending validation.
Advanced Tasks
After integrating real identity and persistence services, retest resource authorization; replace keyword retrieval and evaluate recall; after integrating a model, add factual support, citation accuracy, and cost tracking; before adding write tools, add approval, idempotency, and recovery. Use the project template throughout to save evidence.
Further Reading and Usage Boundaries
- GitHub Actions Official Documentation: Extended reading on related concepts and methods, access verified on 2026-10-14. The exercises, workflows, and metrics on this site are independently organized and are not certified by the source organization.
- Learning Path · Hands-On Projects · Downloadable Templates
This page is a basic guide; production operations must be validated against actual versions, authorized environments, and review evidence.