Every architecture decision in this project traces back to a real business problem faced by a specific kind of employer. This is that employer — a second-generation textile retailer in Nagpur, 52 employees, three retail outlets, and an HR system held together by WhatsApp and a Sunday evening spreadsheet.
A composite profile representing the archetype of the Indian small retail employer — large enough to need HR infrastructure, small enough that every rupee spent on overhead is felt directly in margin.
"I don't need Workday. I need something that works for Ramu, who speaks Telugu and has never opened a laptop in his life. If it runs on WhatsApp, he'll use it. If it doesn't, he won't."Priya Rathi · Owner, Rathi Textiles Pvt. Ltd. · Fictional composite profile
In a 52-person business, the owner is the HR department. Understanding Priya's constraints, goals, and decision-making patterns is what separates an architecture that gets deployed from one that gets demoed and shelved.
The workforce is not monolithic. Three distinct personas represent the range of digital literacy, language, and interaction patterns on the Rathi Textiles floor. The system must serve all three without differentiation.
A system that works for the worker but fails the accountant, or works for Priya but cannot survive a labour inspection, is not production-grade. Every stakeholder's concern is a design constraint.
Before designing an AI system, you audit the client's readiness across four dimensions. Each dimension informs an architectural decision. A low score is not a blocker — it is a design constraint.
Derived directly from the pain points and stakeholder needs above. Prioritised by frequency, compliance risk, and owner time impact. Phase 1 covers the highest-frequency, highest-pain use cases.
| Use case | Actor | Frequency | Current state | Priority | Phase |
|---|---|---|---|---|---|
| Leave application (voice or text) | Any worker | ~40 / month | WhatsApp to supervisor, unrecorded | P1 — Critical | Phase 1 |
| Leave balance query | Any worker | ~25 / month | Ask supervisor verbally | P1 — Critical | Phase 1 |
| Leave approval / denial with policy cite | System → Priya (HITL) | ~40 / month | Priya decides via WhatsApp, no record | P1 — Critical | Phase 1 |
| Policy question (entitlements, rules) | Any worker | ~15 / month | Ask supervisor, often wrong answer | P1 — High | Phase 1 |
| Employee onboarding (doc collection) | New joinee | ~3 / month | Paper forms, frequently incomplete | P1 — High | Phase 2 |
| ESIC / EPF registration trigger | System → CA | ~3 / month | Manual, frequently delayed | P1 — High | Phase 2 |
| Attendance check-in / check-out | Any worker | ~1,100 / month | Register sign-in, often missed | P2 — Medium | Phase 3 |
| Payroll calculation | System → Priya → CA | 1 / month | 4-hour Sunday manual process | P2 — Medium | Phase 3 |
| Payslip delivery via WhatsApp | System → Worker | 52 / month | Not done — workers don't receive payslips | P2 — Medium | Phase 3 |
| Grievance logging and tracking | Any worker | ~2 / month | Verbal, unrecorded, unresolved | P2 — Medium | Phase 4 |
| Labour inspection report export | Priya → CA → Inspector | 1–2 / year | Manual compilation, error-prone | P2 — Medium | Phase 4 |
| Attrition risk scoring | System → Priya | 1 / month | Not done | P3 — Future | Phase 5 |
| Leave pattern anomaly detection | System → Priya | Continuous | Not done | P3 — Future | Phase 5 |
| Workforce demand forecasting | System → Priya | Weekly | Priya's intuition | P3 — Future | Phase 5 |