The challenge
Last-mile dispatch decisions were getting made on regional control-tower whiteboards, and field-service coordination across the network depended on phone calls and ad-hoc WhatsApp groups. On-time performance varied widely by city, and forecasting reaction time to demand shocks was too slow.
How we approached it
Ajuni wired Nila to the Mahindra OMS and the live telematics feed to drive sub-second dispatch routing, and Pragna landed on the field-coordination surface for shift planning. The model layer ran on Azure India with edge caching near the larger control towers to keep latency tight.
Outcomes in production
- 23% lift in on-time performance across served lanes
- Sub-second dispatch routing decisions in production
- Field-coordination shifted off WhatsApp into agents
- Shock-response cadence cut from hours to minutes
- 6 weeks kickoff to live dispatch traffic
Stack & guardrails
Integration & deployment
- Azure India · production tier
- Mahindra OMS integration
- Telematics feed connector
- Claude Sonnet 4 · API
- Edge cache · control towers
- Power BI mirror
Compliance & audit
- DPDP Act 2023
- ISO 27001
- Internal SLA audit
Timeline
-
Week 0–1
Lane + tower scoping
Highest-variance lanes selected with the control-tower operations team.
-
Week 2–3
Telematics + OMS wiring
Live feeds integrated; latency budget validated end to end.
-
Week 4–5
Shadow dispatch
Nila ran parallel to controllers; routing decisions reviewed daily.
-
Week 6
Production cutover
Live dispatch routed through agents; OTP tracked against baseline.
