Notes on building with AI automation
Agents, infrastructure, and direct implementation blueprints, written for Indian SMBs and builders who ship.
Multilingual Hindi Voice Agent Setup: Sarvam AI & n8n
Building a multilingual Hindi voice agent with Sarvam AI and n8n takes just five core components. You configure an n8n webhook for incoming audio, pass it to Sarvam's Saaras STT API, and process it with an LLM. Then, generate spoken Hindi via the Bulbul v3 TTS API and return the base64 audio stream to the caller.
Build an Inbound Exotel AI Voice Agent with n8n | VectorBloom
An inbound AI voice agent pairs Exotel's Mumbai telephony with real-time WebSockets and n8n. Incoming calls stream bidirectional 8 kHz PCM audio to a voice runtime. The runtime handles conversational intelligence. Meanwhile, n8n manages mid-call tool execution, CRM updates, and downstream messaging over webhooks.
Building a Production WhatsApp AI Customer Support Agent with DeepSeek & Gemini via n8n
Building a production-grade WhatsApp AI customer support agent requires connecting WhatsApp Cloud API webhooks to self-hosted n8n, orchestrating context retrieval via vector store embeddings, querying DeepSeek or Gemini 2.5 Flash for deterministic responses, and implementing strict human-handoff triggers when confidence drops below 85%. Operating costs average under ₹0.08 per AI resolution.
Get new posts by email
Direct implementation blueprints for Indian SMBs. One email per post, no spam.