Case study 01 / Workflow orchestration
Orchestrating patient communication.
Calls, waits, messages, and human review, coordinated by one workflow engine. Built from the first commit to production.
- Company
- Quadrivia AI
- Project date
- 2026
- Focus
- Workflow orchestration
Conceptual workflow: a call leads to a wait, then a message or human review. Retry and resume support recovery.
01 / Context
The challenge
A patient journey is more than a single call. It can involve waiting, sending an SMS, taking a different path, or handing a decision to a person. The service needed to coordinate those steps and support recovery when work failed.
02 / Ownership
My contribution
Architecture and implementation, from the first commit through production and subsequent recovery features.
- Architecting and implementing the workflow engine
- Supporting conditional routing and human review gates
- Delivering individual and bulk retries, resume and automatic retry on recovery
- Adding dedicated task queues, dead-letter handling and operational monitoring
03 / Engineering
The approach
Make the journey explicit
I built a directed acyclic graph (DAG) engine to represent multi-step journeys, with conditional routing and human review gates.
Design for the next attempt
Individual and bulk retries, resume, and automatic retry on recovery extended the engine beyond the initial execution path.
Make operations part of the system
Dedicated task queues, dead-letter handling, and monitoring supported the work of running the service in production.
04 / Delivery
What it enabled
The engine went from its first commit to production, coordinating multi-step patient communication with explicit routing, review, and recovery capabilities.
The recovery path deserves as much attention as the first successful run.
Tools & technologies
- Python
- FastAPI
- GCP
- Cloud Tasks
- Pub/Sub
- Grafana
- Loki