Creator websites
Let audiences search your long-form engineering archive.
Embedded search proof
Lightweight embeds that surface timestamped expert moments — not a broad AI answer box.
Same retrieval UX as the main product — embedded mode is in discovery.
<iframe src="https://www.youtubetimesearch.com/embed/search?q=rag+vs+mcp" title="Expert engineering clip search" width="100%" height="420" loading="lazy" style="border:0;border-radius:12px;" ></iframe>
Existing embed routes: /embed/search
Let audiences search your long-form engineering archive.
Jump students to the best explanation inside each lecture.
Make hundreds of hours searchable without manual timestamps.
Embed concept search next to syllabus content.
Trust product layer for agentic rag — cluster rag-retrieval
Why this answer won
Tier-1 expert moment (Anthropic) paired with hard doc citations; fusion=rrf and top_k=20 in operational excerpt.
Configs used
corpus · confidence 43%
Benchmark evidence
faithfulness=0.82
observed in cited evidence
Sam Witteveen
Failure fixes
No incident remediation path in cited evidence.
Expert video corroboration
Building with MCP and Claude API — tool invocation
Sam Witteveen
https://www.youtube.com/watch?v=aZLr962R6Ag&t=638Contradictory evidence
No contradictory expert framing detected.
Trace lineage
queryretrieval.request
hybrid_search
agentic rag
retrieve_hit_1retrieval.candidate
Sam Witteveen
12:34 · score 0.82
synthesisanswer.operational_gate
mcp_tool_orchestration
passed
Citation quality (primary)
MCP is a protocol for tool context, while RAG retrieves documents at answer time. trace id visible in LangSmith retrieve span; faithfulness=0.82.
Authority 82% · high confidence
Winning evidence
Rejected evidence
Operational checklist
Uncertainty