Attendance Without Smartphones
Track daily attendance of 1,000 employees across 100 sites using only the channels available without a personal smartphone: voice, IVR, SMS, supervisor confirmation, and shared site hardware. Every channel feeds the same ledger; the LLM is the reasoning layer that makes the messy multi-source reality usable for HR.
1. Employee
No smartphone, no app
2. Local phone / landline / kiosk
Voice, IVR, SMS, biometric, RFID
3. Location identifier
Each site has a stable id
4. Attendance gateway
Ingests all channels
5. LLM reconciliation
Reasons across sources, flags anomalies
6. Centralised attendance ledger
One source of truth, per-employee
7. HR dashboard
Real-time, plain-English summary
Outbound AI voice call
~70% after rolloutHunar places a 10–20 second call at shift start. The LLM does speaker verification, anti-spoofing, intent classification. Default channel — the system calls the employee, not the other way around.
Where: Every site, every employee with a phone number on file.
Biometric / RFID kiosk
~20% (high-headcount sites)Fingerprint or face kiosk at factory gates; RFID badge readers in warehouses. Zero employee action, hard-to-forge identity. Spend hardware budget on the 20% of sites that hold 80% of headcount.
Where: Large factories, secure sites, warehouses.
SMS / IVR pull channel
~8%Free inbound IVR number (punch in employee ID + voice OTP) and structured SMS (`P 12 4` = present 12, absent 4). LLM parses free-form replies into structured rows.
Where: Small sites, late arrivals, sick leave notifications.
Supervisor fallback
~2%Supervisor confirms shift attendance from a shared tablet or laptop (company device, not personal). Queues offline and uploads when signal returns. The LLM reconciles supervisor confirmations against individual check-ins.
Where: Remote sites, edge cases, dispute resolution.
- A per-employee verdict: present / absent / late / leave / unverified, with a confidence score.
- A short, plain-English summary for HR — '988 present, 9 absent (4 with approved leave), 3 unverified. Exception: Aarav Sharma — voice said present from a number not on file, no biometric match at the gate, supervisor marked absent. Recommend manual review.'
- A weekly payroll-ready export that handles partial days, late arrivals, overtime, and the inevitable 'I was there but the system didn't see me' cases.
- Anomaly flags: buddy-punching, geo-impossibility, pattern anomalies (always-late-Monday, etc.).
Collect only what's needed. Store short summaries, not full audio — unless an exception requires drilling in.
Explain every decision. Click a row, see the chain of evidence: which channel, which timestamp, which confidence.
Opt-in with a real alternative. IVR/SMS path is no-record-by-default. People are not penalised for choosing the lower-fidelity channel.
Week 1
Voice channel
Stand up outbound AI calling — a thin wrapper around the same Hunar integration used in the hiring product. One campaign per day, one question per employee.
Week 2
SMS + supervisor fallback
Add the inbound IVR number and the supervisor SMS path. Most 'missed call' cases get resolved here.
Week 3
Hardware pilot
Buy 5–10 biometric kiosks, deploy at the largest sites. Confirm they work, confirm the data flows, confirm the LLM reconciliation handles the new source.
Week 4
Reconciliation + reporting
Turn on the LLM reconciliation pass. Ship the HR dashboard with the plain-English daily summary and the per-employee audit trail.