M05 Iterative Improvement

Due Date: Anytime til April 27, 2026

Building on the initial evaluation, students will iteratively refine the DAIS system - through architectural or agent modifications - and conduct an ablation study comparing different approaches, ultimately documenting these changes and their impact in a concise iteration report.

Goals:

  • Improved system (architecture or prompt/agent changes) motivated by Milestone 4 analysis.
  • New evaluation results and at least one ablation study (e.g., graph vs vector vs hybrid).
  • Short iteration report summarizing changes and impact.

Deliverables: 0. Create a document M05_MILESTONE.md where you briefly describe parts of your project according to the directions below. (Use the exact filename, it’s case sensitive).

  1. Commit your code and the file M05_MILESTONE.md into your project repository.
  2. Merge to your branch uat. (You don’t need to notify the instructor or add them as reviewer.) This is your review version that will be evaluated. Meanwhile you may continue on your development branch.

Milestone Document:

GoalRequirements
System Refinements ImplementationDescribe the architectural or agent-level modifications made. Explain how each change is linked to the improvement strategies proposed in M04. Include instructions on how to run the updated system.
Ablation StudyDescribe the ablation study design: What alternative approaches were compared (e.g., different retrieval strategies, agent configurations, or prompt designs)? How were results measured against the M04 baseline?
Comparative Results & Impact AssessmentPresent re-evaluation results alongside M04 baseline metrics. Interpret the magnitude of improvements and note any regressions or trade-offs.
Iteration ReportProvide a concise report demonstrating how the performance of the DAIS has improved based on the AI evaluation metrics. Document what was changed, the rationale, and the measured impact on evaluation results.

Evaluation:

#CriterionDescriptionPoints
1System Refinements ImplementationArchitectural or agent-level modifications informed by M04 findings are implemented and functional. Changes are clearly linked to the improvement strategies proposed in M04.50
2Ablation StudyA structured ablation study compares at least two alternative approaches (e.g., different retrieval strategies, agent configurations, or prompt designs), with results measured against the M04 baseline using the same evaluation pipeline.45
3Comparative Results & Impact AssessmentRe-evaluation results are presented alongside M04 baseline metrics in a structured comparison. The analysis interprets the magnitude and significance of improvements and notes any regressions or trade-offs.40
4Iteration ReportA concise iteration report demonstrates how the performance of the DAIS has improved based on the AI evaluation metrics. The report documents what was changed, the rationale, and the measured impact on evaluation results.25