// case study
EVIA
10-agent therapeutic AI platform with layered safety gating and a routing hierarchy. Engineered from Chennai.
Vertex AIGCPmulti-agent orchestrationPython
Problem
01
Build a therapeutic conversational platform where safety is non-negotiable — every user message must pass safety checks before any therapeutic agent acts, and conversations must route to the right specialized agent.
Architecture
02
architecture diagram
drop SVG/PNG here
What I built
03
- 01A 10-agent architecture on GCP / Vertex AI with a dedicated safety-gating layer (A1) screening input before downstream agents act.
- 02A routing hierarchy (A8) directing conversations to the appropriate specialized agent.
- 03Built for a use case where incorrect handling carries real risk — so the safety layer is a hard gate, not a soft prompt.
Results
04
10
Coordinated agents with safety gating as a mandatory first pass
Hard gate
Safety layer — not a soft prompt — on every turn
GCP
Deployed on managed Vertex AI infrastructure
Stack
05
Vertex AIGCPmulti-agent orchestrationPython
Links