ShardaCare: Symptom-to-Specialist Routing in Under 2 Minutes

Key outcomes

• 98% routing accuracy validated against 12,000 verified clinical pathway records.
• Average triage time reduced from 18 minutes (human receptionist) to under 2 minutes.
• 71% reduction in triage operational cost over a 6-month period.
• Zero instances of high-severity mis-triage in post-deployment audit.

The problem

ShardaCare's receptionists were fielding 400+ patient inquiries daily. Most were simple routing decisions — "I have chest pain, who do I see?" — but required trained staff to handle safely. Wait times for triage stretched to 20 minutes during peak hours, causing patients to drop off before they got help.

The Reinforce approach

1. 01
   Clinical pathway dataShardaCare's medical team provided 12,000 annotated symptom-to-specialist mappings. Reinforce structured these as a preference dataset for RLHF fine-tuning.
2. 02
   Safety guardrailsA hard-coded escalation layer detects red-flag symptoms (chest pain + breathlessness, sudden vision loss, etc.) and immediately routes to emergency — bypassing the AI decision entirely.
3. 03
   Privacy-first architectureAll patient context is encrypted at rest and in transit. The model runs within ShardaCare's own cloud tenant — no patient data leaves their environment.
4. 04
   Continuous evalWeekly routing accuracy reports generated from doctor feedback ("wrong department" flags) feed back into the training loop automatically.

“Bhartiya Didi handles the triage layer with the accuracy of a trained nurse and the speed of a search engine. Our doctors now spend their time on diagnosis, not reception.”