I’m trying to get this flow consistently fast (aiming for sub-10s per run).
Current setup:
- Webhook → Code (preprocess)
- AI Agent (OpenAI Chat) with Pinecone Vector Store + OpenAI Embeddings
- BuildPayload → ElevenLabs (TTS) → rename → Supabase Storage upload
- In parallel: Supabase DB update + Analytics (Mixpanel)
- Final Respond to Webhook returning audio URL
Questions for the pros:
- Best way to parallelize TTS/upload/analytics—Queue Mode + workers or sub-workflows?
- Any wins from HTTP keep-alive, batching, or reducing hops (replace multiple Set/If with one Function)?
- Tips for caching/pre-warming (tokens, Pinecone lookups) and passing binary data to avoid base64 bloat?
- Anyone using streaming (LLM partials → start TTS early) to overlap steps?
- Execution-data/DB settings you tweak to lower overhead?
Short node patterns or screenshots would be awesome—thanks! 🙏