Skip to content

Latest commit

 

History

History
48 lines (29 loc) · 2.38 KB

demo-script.md

File metadata and controls

48 lines (29 loc) · 2.38 KB

Three-minute demo script

0:00–0:30 — Frame the problem

Open the landing page. Explain that a production support Agent needs more than a chat box: it must retrieve policy, inspect live business state, expose safe execution evidence, and stop before sending risky answers.

0:30–1:20 — Customer workflow

  1. Open Knowledge and upload seed/knowledge/subscription-activation.md.
  2. Open Customer Portal.
  3. Ask: My Pro payment succeeded but workspace is still Free.
  4. Point out the visible stages, three read-only tool calls, cited policy excerpt, and generated answer.
  5. Show that the P1 guard does not send the draft. Create a Ticket.

1:20–2:05 — Human review

  1. Open Agent Workspace.
  2. Select the P1 Ticket.
  3. Show intent, confidence, citations, and business facts.
  4. Click 批准并发送 and explain the confirmation gate.
  5. Confirm approval and show 已解决.
  6. Mention that review, final message, Ticket state, and Run state are one transaction.

2:05–2:35 — Observability and replay

Open a Run detail URL. Show ordered steps, durations, safe tool arguments, citations, and error codes. Explain recursive secret redaction and that replay creates a new linked Run without changing history.

2:35–3:00 — Evaluation and trade-offs

Open Agent Evaluation and run the Mock suite. State clearly that 24/24 is a deterministic contract check, not a model-quality benchmark. Highlight stored dataset hash, commit SHA, mode, timestamp, numerator/denominator metrics, and per-case failures.

Common interview follow-ups

Why Go and Eino?

Go makes concurrency, typed boundaries, and small deployable services straightforward. Eino supplies a composable Agent runtime while the project keeps domain rules and observability outside model prompts.

Why a separate MCP server?

It proves protocol-level tool integration and keeps the business boundary independently testable. The tools are intentionally read-only.

How do you reduce hallucination?

Hybrid retrieval, typed business facts, citation requirements, output schema validation, confidence/risk gates, and human review for P0/P1 or weak evidence.

What would change in production?

Replace demo headers with OIDC, add tenant isolation and rate limiting, use a managed vector store/observability backend as scale requires, and establish real-model evaluation thresholds before automated sending.