Disclaimer: I created this piece of technical content for the purposes of entering the Google “All Things Agentic Hackathon” (August 2026).
The Reality of Modern Business Communication
In the fast-paced digital economy of Southeast Asia and beyond, small and medium enterprises (SMBs) face a massive operational challenge: fragmentation and response latency.
A typical merchant manages customer inquiries across five distinct channels every day: WhatsApp Business, Instagram Direct Messages, TikTok Shop Chat, Facebook Messenger, and Web Chat widgets. When high-intent buyers message a business at 3 AM or during peak rush hours, human customer support teams simply cannot keep up. Research shows that over 40% of prospective buyers abandon their carts or move to competitors if their inquiries are not answered within 5 minutes.
Traditional chatbots have failed to solve this. Rule-based bots feel robotic, break easily on informal slang, and cannot take dynamic actions. Meanwhile, simple “prompt wrapper” AI models can talk nicely, but they lack real-time business context and cannot execute autonomous backend workflows.
To solve this, we engineered Tanyaiku — a multi-tenant Autonomous AI Agentic Growth Engine built natively on Google Cloud and Gemini.
The Core Architecture: From Passive Chat to Autonomous Action
Tanyaiku is designed not just to answer questions, but to take multi-step actions across an enterprise’s customer journey.
1. Dynamic Inventory Memory Bank (Context Override)
In real-world retail, products run out of stock unexpectedly. Instead of requiring developers to retrain models or edit complex databases, Tanyaiku provides an instant Memory Bank Override. When a merchant toggles a product to “Out of Stock”, the Gemini agentic pipeline immediately injects this constraint into its short-term reasoning memory, politely declines out-of-stock orders across all 5 channels, and suggests matching alternatives in real-time.
2. Autonomous Lead Intent Scoring (CRM)
Every inbound conversation is evaluated asynchronously by our agent execution loop. The agent classifies the buyer’s purchase urgency into Hot, Warm, or Cold leads. For high-ticket inquiries or urgent bulk orders (Hot Leads), the agent triggers an immediate alert to the business owner’s private WhatsApp.
3. 24/7 Outbound Cart Recovery Agent
Tanyaiku runs an asynchronous background loop on Google Cloud Run that monitors abandoned checkout interactions. It intelligently schedules tailored follow-up messages and personalized incentives after a set cool-down period to recover lost sales.
Google Cloud Infrastructure & Tech Stack
- AI Model & Reasoning: Google Gemini 3.6 Flash via the modern
@google/genaiTypeScript SDK. - Backend Runtime: Node.js / Express.js containerized and deployed on Google Cloud Run (
asia-southeast1). - State & Data Store: Google Cloud Firestore providing ultra-low-latency real-time data persistence and multi-tenant isolation.
- Frontend Dashboard: React 18 with Tailwind CSS.
Key Takeaways & What’s Next
Building with the Gemini SDK and Google Cloud Run allowed us to achieve sub-second response times while running complex background agentic tasks without blocking the customer conversational stream.
Our mission with Tanyaiku is to democratize autonomous AI agents for businesses of all sizes, turning chaotic customer service inboxes into automated revenue growth engines.
Project Links:
- Live Platform Preview: Tanyaiku on Cloud Run
- Built by: Akhmad Khudri (Lead Architect) in collaboration with PT Konakami Digital Indonesia.
