Two AI systems for a full patient journey
Manager of Technology & Product · Sirka (YC S21) · 2026
Designed and shipped two production AI systems covering the intake-to-consult journey, validated in a controlled clinical pilot with no adverse events — and no added clinical headcount.
Context & Constraint
Growing lead volume, limited clinical capacity, and heavy drop-off before consult. One non-negotiable rule: the physician always performs diagnosis and prescribing — never the AI.
Decision + Tradeoff
I split the system in two rather than building one large agent: a Receptionist AI on WhatsApp for intake and triage, and a Doctor AI for pre-diagnosis support handed to the physician. The split kept every AI action on the assist side of the clinical line — clearer accountability and safety, at the cost of more integration surface. I chose that over a single monolithic agent because in regulated care, a defensible boundary beats a slightly slicker flow.
Outcome
- Consult start time cut dramatically
- Physician time per patient cut significantly
- Consult conversion improved substantially versus baseline
- Validated in a controlled clinical pilot with no adverse events
- Admin team redeployed from manual intake to higher-value follow-up — no redundancies