AWS case study: Social Department's QuickSearch finds memeable moments across video libraries
AWS ML Blog · rss · 2026-11-19
AWS ML Blog details how Social Department built QuickSearch, a video discovery system for entertainment marketing that answers judgment-based queries (shareable scenes, quotable quotes, memeable frames, trend alignment) rather than object-level semantic search. Architecture: a content understanding pipeline (S3 → EventBridge → Step Functions orchestrating 24 Lambdas → MediaConvert → Rekognition shot detection), independently scheduled trend acquisition lanes, and a QuickSearch job layer joining both at request time. Video reasoning runs on TwelveLabs Marengo 3.0/Pegasus 1.2 on Amazon Bedrock, retrieval via OpenSearch Serverless, plus Claude models. Guardrails include code-based grounding checks, Bedrock Guardrails for prompt attacks, and human approval before publishing — "AI proposes, people decide."
More from coding & agent
- Open-source MCP server controls Logitech Harmony hubs locally with LLM-safe guardrails — SDNick484 · 2026-11-19
- Study: given an empty computer, strong models write their own games to self-improve — CatAstro_Piyush · 2026-11-19
- Elicit on agent swarms: pharma winners of 2036 will robustly deploy them without going rogue — jungofthewon · 2026-11-19
- Hands-on GPT 6.2 review: Astra and Sol are great, but no drop-in replacement for Fable 5.2 or Opus 5.5 — kieranklaassen · 2026-11-19
- Suno MCP now open to all: author makes a full music video via Claude lyrics and Claude Code editing — minchoi · 2026-11-19
- PromptMaker: open-source local prompt studio for SD workflows, runs on 8 GB GPUs — ChanceLingonberry170 · 2026-11-19