Technical Architect
Skor Insights
A document-forensics and tenant-risk platform combining LLM analysis with deterministic fraud checks.
- Delivery scope
- Application to infrastructure
- Processing
- Async and elastic
- Client team
- 6-7 stakeholders
01 · Context
The system behind the product
Owned the platform architecture, document-forensics engine, infrastructure, and technical delivery for a Spain-based proptech team.
02 · Architecture
How the system is shaped
- 01
A Python and FastAPI forensics service combines deterministic OpenCV and PyMuPDF checks with LLM-assisted document analysis.
- 02
RabbitMQ and Celery isolate long-running report work from request handling, with MongoDB preserving job and report state.
- 03
Azure Container Apps and KEDA scale workers against queue pressure while the MERN application remains independently deployable.
03 · Decisions
Trade-offs made explicit
Evidence before inference
Deterministic document signals remain explicit and reviewable; LLM analysis adds context without becoming the sole fraud verdict.
Queue-first processing
Report generation is treated as asynchronous work so traffic spikes do not turn into request timeouts or oversized always-on infrastructure.
Independent scaling boundaries
Frontend, API, forensic workers, and data services can evolve and scale according to their own load profiles.
04 · Outcome
What changed
Created an elastic asynchronous processing system for variable report loads and a risk-scoring workflow used in rental verification.
- Created a repeatable tenant-risk workflow from uploaded documents through forensic analysis and report generation.
- Made variable report-processing demand operationally manageable through queue depth-driven autoscaling.
- Established a shared architecture across the client team, MERN developers, Python services, data layer, and infrastructure.
05 · Capabilities
What the work involved
- Document fraud detection
- LLM-assisted analysis
- Asynchronous processing
- Engineering leadership