WhatsApp CS Bot — BULOG
An AI customer-service bot on WhatsApp for the BULOG PBP program: a 41+ item FAQ knowledge base, automatic media replies, scheduled broadcasts, and field-officer reports that land straight into Google Sheets — all orchestrated by the Hermes agent framework.
// Overview
The PBP (Program Bantuan Pangan) program generates a steady stream of the same questions from beneficiaries — eligibility, schedules, locations, requirements. I built a WhatsApp bot that answers them automatically from a curated knowledge base, with the escalation path and administrative reporting built in.
The bot runs on Hermes, my multi-profile AI agent orchestration, with LLM provider failover (MiMo, DeepSeek, DeepInfra) so the service survives a provider outage — the same 24/7 resilience approach I use for my network infrastructure.
// The Challenge
Officers spent their day answering repetitive questions manually, broadcasts went out by copy-paste, and reports from field officers arrived as scattered chat messages that had to be re-typed into spreadsheets. The bot had to be accurate on program rules (not hallucinated), safe for use by the general public, and useful for three very different user groups at once.
// What I Built
An orchestrated WhatsApp agent with guardrails and real workflow integrations:
Curated Knowledge Base
41+ FAQ items grounded in actual program rules — answers come from the knowledge base, not free-form generation.
3-Tier User Access
Admin, field officer, and general-user levels — each tier gets different commands and different data visibility.
Officer Reports → Sheets
Field officers file reports through the bot; entries flow automatically into Google Sheets — no re-typing.
Scheduled Broadcasts
Cron-scheduled broadcasts with media auto-replies for program announcements and documentation.
Provider Failover
Multi-provider LLM routing (MiMo, DeepSeek, DeepInfra) with rate limiting — answers keep flowing even when a provider is down.
Persistent Memory
Conversation context persisted via Obsidian-backed storage under the Hermes multi-profile architecture.
// Results
The bot handles the program's routine question load automatically, broadcasts run on schedule, and officer reporting moved from scattered chats into a structured spreadsheet pipeline — all on infrastructure that has stayed up 24/7.
41+
FAQ knowledge base items
3
User tiers (admin / officer / public)
24/7
Uptime with provider failover
0
Manual re-typing of officer reports
// Tech Stack
Need a bot that actually knows your program?
I design AI agents with curated knowledge bases, access tiers, and real workflow integrations.
Get in Touch