Sentinel AI is being developed as an enterprise-grade AI digital media forensics platform with architecture and operational concerns treated as first-class requirements.
SYSTEM MAPAPI → persistence → intelligence
ClientAPI consumer
FastAPIHTTP boundary
ApplicationTyped domain logic
SQLAlchemyPersistence layer
AI / MLExtensible analysis
PostgreSQLDurable data
Current backend foundation and intended extension boundaries for Sentinel AI.
The problem
Digital media analysis can require multiple AI capabilities, persistent evidence, asynchronous processing and clear auditability. The platform needs an architecture that can grow without turning model experimentation into an unmaintainable backend.
Engineering challenges
Establishing boundaries before implementation grows.
Designing persistence around structured forensic data.
Creating a backend foundation suitable for asynchronous AI workloads.
Keeping the repository maintainable as additional forensic capabilities are introduced.
Architecture-first backend
The backend is centered on FastAPI with typed application boundaries and SQLAlchemy-backed persistence. PostgreSQL provides the relational foundation for durable application data.
The architecture is intentionally designed to accommodate background jobs, storage, vector search and AI/ML models as the platform evolves.
A forensic platform can quickly accumulate models and endpoints without a stable foundation.
Decision
Define repository structure and architectural boundaries before expanding functionality.
Rationale
Early architectural discipline reduces coupling and makes future capabilities easier to integrate.
Trade-offs
Initial development is slower than rapidly adding isolated endpoints.
Some abstractions are designed before all future workloads are known.
02
PostgreSQL as the relational foundation
Context
Forensic workflows need durable, queryable structured data.
Decision
Use PostgreSQL through SQLAlchemy.
Rationale
A mature relational database provides strong consistency, expressive querying and a clear persistence model for core application data.
Backend foundation
The repository is being built with modern Python tooling, strict typing, FastAPI, SQLAlchemy and migration-ready persistence.
Implementation is intentionally incremental so each architectural layer can be validated before the next layer is introduced.
Production mindset
Reliability is being addressed through explicit boundaries, validation, migrations, typed code and a roadmap toward background processing rather than placing every workload directly inside request handling.
Forensic context
The platform is designed around evidence-oriented processing, which makes traceability, controlled data handling and explicit service boundaries important architectural concerns.
Lessons learned
Architecture-first development is most valuable when a system is expected to evolve for years.
AI systems still need conventional software engineering discipline around APIs, persistence and operations.
A clean foundation makes experimentation safer rather than slower in the long run.
What comes next
Introduce asynchronous processing for expensive analysis.
Integrate object storage and vector search where justified.
Add model orchestration and forensic evidence workflows.
Deploy the platform through a production cloud architecture.
