Naouel Eldjouher

Data layer done. Next: the FastAPI application, authentication, and the first real routes.

This week I built the database layer for OmniDomain's FastAPI backend. Three files, each with one job. Moving from direct SQL in Streamlit to a production-grade session, ORM, and repository architecture.

The original project had Streamlit talking directly to PostgreSQL and AWS Batch. That was a deliberate choice. I wanted to focus on the pipeline infrastructure.

Same data, one extra flag: --reference. The pipeline switched from Helixer to BRAKER3 automatically. 770 proteins vs 527 — a 46% increase from protein evidence alone.

With the organelle case validated, PhytoFlow moved to its first nuclear genome run. The goal was not a publishable genome, it was a pipeline validation:

OmniDomain now has three new pipelines under active development: PhytoFlow for plant genome assembly and annotation, FungalFlow for fungal genome assembly and biosynthetic gene cluster discovery, and MetaCflow for bacterial shotgun metagenomics and MAG recovery.

Three pipelines. One repo. The structure determines everything. Each pipeline gets its own main.nf and modules.config. One global nextflow.config handles profiles, resource limits and shared params.

After finishing NextAMR, I'm expanding to fungi and plants. OmniDomain is a unified Nextflow + AWS Batch platform for bacterial, fungal, and plant genomics — one shared infrastructure, three pipeline engines.