DmitryDmitriadi/multi-agent-workflow-platform

A multi-agent platform for running real-time conversational video avatars that conduct structured sales/pre-sales dialogues (greeting → diagnosis → objection handling → CTA) with sub-second per-turn latency. Intended for teams building live avatar assistants and integrations with real-time CVI APIs. Key features: - Orchestrator that runs a typed stage graph; transitions driven by structured signals (intent classifier, counts, explicit requests), not free-form LLM reasoning. - Specialized agents: knowledge (RAG retrieval + source tags), objection classifier/strategies, analytics, avatar/voice lifecycle, reviewer (policy gate on every outbound utterance), cost guardian (per-session budgets, graceful wrap-up), safety layer. - Per-turn knowledge grounding with source attribution, structured outputs at every agent boundary, OpenAI-compatible endpoint, mock mode for development, multi-language support, and session lifecycle/webhook handling.

このRepoのdescription / GitHub topics / READMEのAI要約に出現したAI関連キーワードをカテゴリ別に表示。各バッジは該当カテゴリの詳細ランキングへリンクします。