Shiftly
// Multi-Agent AI-Assisted Platform for Rotating Shift Optimization
Supervisors need a fast, flexible way to build rotating shifts without committing legal errors, while employees need predictability and accessible channels to check and manage their schedules.
DISCOVERY & RESEARCH PHASE
The Supervisor Pain Point: Manual Friction & Last-Minute Scrambles
Building the weekly schedule takes between 6 and 10 hours per week in Excel. Any unexpected absence disrupts the entire plan, triggering frantic off-hours phone calls.
The Employee Pain Point: Lack of Notice & Bureaucratic Swaps
Frontline workers receive schedules with short notice (often via photos of printed sheets on bulletin boards). Requesting a shift swap with a coworker involves days of bureaucratic friction.
The HR Department Pain Point: Unintentional Labor Violations
Failing to enforce mandatory 12-hour minimum rest periods between shifts exposes the company to hefty labor fines and legal claims.
PROBLEM FRAMING & PRODUCT KPIS
“Supervisors need a fast, flexible way to build rotating shifts without committing legal errors, while employees need predictability and accessible channels to check and manage their schedules.”
Needs to build schedules considering availability, leaves, and staffing demands without spending his entire weekend aligning Excel cells.
Works on the shop floor, does not use a computer during his workday, and needs to access his schedule through a channel he already uses daily (Mobile App / WhatsApp).
Reduce shift generation time from 8 hours to < 5 minutes.
0% labor law violations regarding minimum rest periods in published schedules.
90% active employee engagement via Mobile App/WhatsApp within the first 45 days.
TEST & DISCOVERY EXPERIMENTS
| RISK / HYPOTHESIS | EXPERIMENT | SUCCESS CRITERION |
|---|---|---|
AI Feasibility: Standard LLMs can generate coherent schedule combinations from natural language. | Run the Planner Agent against 20 complex historical scheduling scenarios. | 90% valid and coherent schedules generated on the first attempt. |
Legal Compliance: Auditor Agent detects all non-compliant shift assignments. | Red Teaming: Feed the Auditor Agent 50 schedules containing deliberate labor law violations. | 100% detection rate of hard constraint violations. |
Panel Usability: Supervisors can easily review, edit, and publish AI-proposed rosters. | Usability testing with 5 real supervisors using the web panel prototype. | 4 out of 5 supervisors successfully edited and published schedules in < 30 minutes. |
PRODUCT REQUIREMENTS DOCUMENT (PRD)
| REQ ID | FEATURE AREA | FUNCTIONAL DESCRIPTION | PRIORITY |
|---|---|---|---|
| FR-01 | Multi-Agent Generation | Planner Agent parses natural language constraints and generates schedule proposal. | MVP (P0) |
| FR-02 | Legal Compliance Auditor | Deterministic Python rule engine audits rest periods and union contracts before supervisor review. | MVP (P0) |
| FR-03 | Supervisor Review Panel | Interactive matrix allowing manual cell overrides, conflict highlighting, and final 1-click publish. | MVP (P0) |
| FR-04 | Employee Mobile Schedule | Color-coded shift calendar (Morning, Afternoon, Night, Off) with instant push alerts. | MVP (P0) |
| FR-05 | Peer-to-Peer Shift Swaps | 2-click shift exchange request between coworkers with automated supervisor approval. | MVP (P0) |
| FR-06 | WhatsApp Query Bot | Dialogflow CX bot answering 'When do I work tomorrow?' without incurring LLM token costs. | Phase 2 (P1) |
| FR-07 | Clock-In/Out Integration | API connectors linking actual punch-clock timestamps with scheduled shift rosters. | Phase 2 (P1) |
Schedule generation under 45s for 100+ employees; multi-agent retry backoff under 15s.
Auditor Agent is 100% deterministic (Python rule engine, zero stochastic hallucinations).
RBAC separating Supervisors, HR Admins, and Operational Staff with encrypted employee PII.
PRODUCT ROADMAP & PHASING
Multi-Agent Core & Human-in-the-Loop
- →Supervisor Web Panel with Natural Language Prompting
- →Planner LLM + Python Rule Engine Auditor
- →Employee Mobile App (Shift Calendar & 2-Click Swaps)
- →Push Notification Dispatcher
Conversational AI & Attendance API
- →WhatsApp Chatbot via Dialogflow CX
- →Biometric Clock-in/out Hardware API Connector
- →Overtime & Absence Forecasting Dashboard
Predictive Demand & Enterprise Multi-Branch
- →Foot-Traffic & Sales Volume Demand Auto-Forecasting
- →Multi-Branch Staff Sharing Pool
- →Enterprise SAP
PROTOTYPES, WIREFRAMES, MOCKS & USER FLOWS
Employee View of weekly schedule
Supervisor View of Dashboard
DF flow for checking shifts and offloading calls to AI models through Whatsapp chat
KEY TAKEAWAYS & LESSONS LEARNED
Human-in-the-Loop is Essential in B2B AI: Supervisors do not want an AI that makes blind autonomous decisions; they prefer an AI that handles 95% of the heavy lifting while leaving the final approval button in their hands.
Meeting Users Where They Are Reduces Friction: For frontline staff, forcing a native app download can be a high adoption barrier. Including WhatsApp in the product vision guarantees day-one engagement.
Balancing Flexibility with Control: Combining LLMs (to understand natural language inputs) with deterministic code validators (for legal auditing) proved to be the winning formula to mitigate AI hallucination risks.