Production system · Internal useOperational data · Internal production
Service operations intelligence and SLA engine
Project details
Why it was created
Service records arrived through different APIs and files with inconsistent status histories. Simple SLA arithmetic could not represent business hours, holidays, requester pauses, vendor time and reopened work reliably.
What I created
I designed the incremental ingestion, analytical model and SLA rules, then connected the resulting views to the daily routines of Service Desk, field teams and operational leadership.
Constraints and trade-offs
+
Constraints that shaped the system
- 01The model had to preserve synchronization history and expose data-quality failures instead of silently normalizing them away.
- 02SLA clocks changed with priority, national holidays, pauses, vendor ownership and reopened tickets.
- 03Operational and executive users needed different levels of detail from the same trusted dataset.
Trade-offs
- 01Incremental synchronization with durable history was more complex than full reloads, but made changes and gaps explainable.
- 02DuckDB kept analytical queries fast and operationally simple for an internal product without prematurely adopting a distributed warehouse.
- 03Controlled exports favored traceability over unrestricted spreadsheet copies disconnected from the source model.
Additional architecture decisions
- Incremental API and file ingestion normalizes more than 32,000 records into an analytical model with synchronization history and quality checks.
- The SLA domain handles priority policies, business hours, national holidays, requester pauses, vendor time and reopened work.
- FastAPI and DuckDB serve React dashboards with heatmaps, period comparison, operational queues and controlled exports.
- Recurring executive and field reports connect the product to real Service Desk, monitoring and critical-incident rituals.
Business value
More than 32,000 service records became a quality-controlled operational model serving SLA, backlog, workload, heatmaps and recurring leadership reports from one source of truth.
Engineering principle
“A dashboard earns trust before it earns attention: people act on a metric only when its clock, exclusions and source history can be explained.”