Prepared for the Agentic AI assignment. Not presented live; carried for Q&A and to cover the brief's "Elite / maximum differentiation" markers (§16). Each file maps to a rubric dimension.
| File | Rubric dimension it defends | Brief §16 marker |
|---|---|---|
why_agent_not_dashboard.md |
Problem Framing (15) | Comparison vs non-agentic alternative |
risk_matrix.md |
Risk Awareness (15) | Risk matrix + approval-tier matrix + failure-mode table |
evaluation_metrics.md |
Benefits/Feasibility | 3-layer evaluation metrics |
qa_cheatsheet.md |
Clarity & Presentation (20) | Per-member Q&A readiness |
sequence_diagram.md |
Agentic Thinking (25) | Sequence diagram / agent trace |
failure_modes.md |
Risk Awareness (15) | Failure-mode table (FMEA) |
confidence_policy.md |
Agentic Thinking (25) | Confidence-threshold policy |
audit_log_schema.md |
Risk / Feasibility | Sample audit-log schema |
RECORDING.md |
Demo | Golden-run recording steps (you run once) |
Already in the repo and counting toward the same goal:
docs/tool_catalog.md— Tools & feasibility (15): every tool's spec + fallback.prompts/— the full 4-type prompt pack (system / task / guardrails / self-eval), live in the runtime.docs/architecture.mmd— architecture diagram (→ Figma for the slide).logs/tos_audit.jsonl— sample agent trace / audit schema (record the golden run).
- Record the golden run (all 3 paths) → demo plays from replay, zero credits.
- Build the ~6 slides (Higgsfield + Figma) from the brief §13 outline; architecture.mmd → slide 4.
- Reframe the tool slide as 5 capability clusters, not 18 tools.
- Rehearse so every member is fluent on the Universal Q&A.