Demo
Demo day: the seven-beat live walkthrough, failing gracefully on purpose, peer review — and the RAG Developer credential.
Final act: a live demo that shows the system, not just lucky answers — then peer review, then the credential. The demo structure below is designed around one principle: showing your failure handling earns more trust than showing your successes.
The seven beats (10 minutes)
- 1The corpus and its users (1 min): what documents, who asks, the two hardest things about this corpus.
- 2A routine answer (1 min): clean question → cited answer → click a citation open. Let them watch the quote match the source.
- 3The trace (2 min): same query through the pipeline log — rewrite, retrieved ids + scores, context, verification. This beat separates you from every thin demo they've seen.
- 4A hard case (2 min): your best multi-chunk or exception-clause question. Show the golden-set entry for it — 'this used to fail; here's the log line for the change that fixed it.'
- 5The deliberate failure (2 min): ask something the corpus can't answer, live. Abstention path, nearest sources, LOW confidence. Then the money line: 'here's the ranked list of questions it declines — that's the docs roadmap.'
- 6The scorecard (1.5 min): headline metrics, the calibration table, one known limit stated without flinching.
- 7What it would take to run for real (30 sec): the runbook exists, the gates exist, here's the weekly loop. End operational, not aspirational.
Peer review & credential
- 1Review one peer's full capstone: design doc, eval report, and a hands-on session with their app — including three questions you invent, one designed to force abstention. Rubric: corpus & ingestion craft / retrieval quality & evidence / trust layer / eval rigor / operational readiness (20% each).
- 2The one-change rule, one last time: incorporate your reviewer's strongest finding before final submission, logged as a version bump with its eval delta.
- 3Submit: design doc, frozen system (repo or archive), eval report, demo recording or live slot, your peer review.
Passing earns RAG Developer (credential format EDOVA-RAG-2026-XXXX): you build production RAG pipelines, evaluate retrieval quality with real metrics, and ship citation-aware applications with calibrated trust. Alumni have deployed their capstones as internal doc-bots within the month — the distance from this course to production is intentionally zero.
Agentic AI Systems is the natural sequel — your RAG pipeline becomes a tool an agent calls mid-task, alongside others. Data-track learners: the Conversational Analytics Agent course applies the same grounding discipline to governed SQL. And when your RAG corpus is a company's warehouse and dashboards, look at our products — Vigil monitors the pipelines your RAG app depends on. Retrieval, it turns out, was the gateway drug to data infrastructure.