Katuwang.ai

The Problem

Small Filipino businesses (salons, resorts, clinics) get flooded with repetitive Messenger inquiries — "How much?", "Where are you located?", "Can I book for Saturday?" — and lose potential customers when they can't respond fast enough. Hiring a dedicated staff member for this is too expensive for micro-businesses.

The Solution

I built Katuwang.ai as a multi-tenant Messenger chatbot that handles common inquiries in Taglish (Tagalog-English mix) with a polite Filipino service tone.

• Context retention across conversation turns so the bot remembers what the customer asked about.
• Multi-tenant architecture — Each business gets its own configuration (pricing, location, services, FAQ) stored in Supabase with Row Level Security.
• Lead capture and handoff — When a conversation requires human attention (custom requests, complaints), the bot captures contact details and alerts the business owner instantly.
• Google Sheets/Calendar sync — Booking confirmations auto-populate the business's scheduling tools.

What Went Wrong

The initial prompt engineering produced responses that were too formal — like a corporate chatbot, not a friendly Filipino staff member. Customers would disengage because the tone felt robotic.

The fix: I rewrote the system prompts with explicit tone examples using natural Taglish patterns and common Filipino customer service phrases. I also added a few-shot context with real conversation samples from partner businesses, which dramatically improved the conversational quality.

Results

• Reduced response latency from hours to seconds for common inquiries
• Lead capture with instant owner alerts improved conversion rates
• Supports multiple businesses on a single deployment via Supabase multi-tenancy