Business Context
A high-growth startup with 100+ remote and field employees was spending too much time on administrative overhead: checking attendance receipts, answering simple policy questions, and routing leave requests.
Problem
The HR team was overwhelmed. 60% of their day was spent on WhatsApp answering the same five questions or verifying PDF screenshots of attendance. This manual work prevented them from focusing on strategic hiring and culture building.
Constraints
- Employees must not need to install a new app (must use WhatsApp).
- System must interpret natural language (Indonesian/Regional slang).
- Real-time integration with the existing order spreadsheet.
Architecture Design
I implemented an AI Agent layer using LLMsand n8n. To handle massive internal policy documentation, I utilized AI Agent and RAG (Retrieval Augmented Generation) with a Pinecone vector database to provide highly accurate, context-aware answers.
Implementation
The core logic was built on n8n for rapid workflow iteration. I used Gemini's Assistant API for memory-persistent conversations. For attendance verification, I implemented a QR Code Scanning module that scans submissions and automatically logs timestamps to the database, flagging any anomalies for human review.
Business Impact
Efficiency
HR administrative workload reduced by 70%.
Response Time
Employee inquiries answered instantly 24/7 (was ~4 hour lag).
Data Integrity
Attendance fraud reduced to zero due to daily reporting and QR code scanning.
Employee Satisfaction
Instant leave approval workflows significantly improved employee NPS.
Technical Stack
- n8n
- Gemini 3 Flash
- Pinecone (Vector DB)
- WhatsApp API (WAHA)
- QR Code Scanning
- Supabase