Case Study
Capstone Project ĀKI
ĀKI was an IT7510 capstone (45 credits, Grade A) focused on delivering a web-based AI “virtual assistant” for timetable and course/IT query support. The work combined LLM system prompt engineering with disciplined reliability evaluation across multiple local runtimes, plus a security-minded testing loop - delivered with structured project management through to final submission.
What I led
As Project Manager, I led a team of 4 students to coordinate planning and delivery across all phases of the capstone: proposal/milestones, research compilation, documentation, demonstrations, and structured testing/verification. I owned coordination workflows to keep the team aligned and the work progressing through the full 16-week duration.
Key deliverables
- Final approved project proposal, objectives, and milestone plan (Agile/Kanban delivery)
- Project timeline and milestones documentation (including Gantt/milestone planning)
- Research compilation documenting open-source LLM hosting environments and tooling experiments
- System prompt/timetable documentation and evaluation assets used during demonstrations
- Demo materials (video walkthroughs) showing AI behaviour across different execution platforms
- Testing/verification documentation (UAT + unit testing + security testing evidence)
- Repository deliverables and final submission documentation for capstone completion
Methodology & tools
- Agile methodology with Kanban workflow to manage discovery, development, testing, and iteration
- Jira for task tracking and sprint/milestone coordination
- Git/GitHub for version control and collaborative development
- Weekly planning and progress communication to keep the team aligned (e.g., Discord/messenger)
- System prompt engineering workflow: defining prompt rules, iterating, and validating behaviour
- Web-based chatbot delivery connected to local model environments for consistent testing/benchmarking
- Structured documentation habits to support reproducibility and final submission quality
Testing & security focus
- System prompt injection and behavioural testing (conflicting prompts and adversarial-style inputs)
- Checks for prompt leakage / system prompt disclosure risks
- Security testing evidence as part of iterative verification (UAT + unit testing)
- Evaluation across multiple runtime environments to compare reliability and output characteristics
- Security hardening concepts such as input validation, rate limiting, and security headers (as part of verification)
Key skills
What this shows
This capstone demonstrates strengths: structured project leadership, the ability to translate academic research into practical engineering outputs, and a security-minded approach to evaluating AI behaviour.