arch

cerebro-reranker/ ├── flake.nix # Entry point ├── flake.lock ├── modules/ │ ├── reranker-service.nix # Core service │ ├── ipfs-cluster.nix # Distributed model serving │ ├── training-pipeline.nix # GCP training automation │ ├── cache-layer.nix # Redis + IPFS cache │ └── monitoring.nix # Observability stack ├── src/ │ ├── reranker/ │ │ ├── server.py # FastAPI server │ │ ├── hybrid_engine.py # Hybrid reranking logic │ │ ├── models.py # Model management │ │ └── cache.py # IPFS-backed caching │ ├── training/ │ │ ├── train.py # Fine-tuning pipeline │ │ └── export.py # ONNX export + quantization │ └── lib/ │ ├── scorer.rs # Rust FFI for fast scoring │ └── ipfs_client.rs # IPFS pinning client ├── configs/ │ ├── models.toml # Model registry │ └── gcp-training.yaml # Vertex AI config └── scripts/ ├── setup-gcp.sh └── deploy.sh