System Architecture

LOCAL

Full-stack research platform running entirely on localhost. No cloud dependencies.

🐍

patz_env (conda)

Python 3.11.9 · PyTorch 2.4.1+cu124 · CUDA 12.4 · fair-esm · captum · mlflow · hydra-core · rdkit

⚙

FastAPI Backend

localhost:8007 · 20+ REST endpoints + WebSocket /ws/system /ws/training · serves static frontend · uvicorn ASGI

⚛

Next.js Frontend

localhost:3007 · Next.js 14.2 · React 18 · TypeScript · Tailwind CSS v3 · Plotly.js · rewrites /api/* → :8007

🗄

Data Layer

data/processed/ — dataset_summary.json, split_manifest.json · results/ — test_results.json, xai_results.json, screening_results.csv

📊

MLflow Tracking

mlruns/0/ — 15 experiment runs · meta.yaml + metrics/ + params/ per run · no MLflow server needed (direct file read)

⚙

Configs (Hydra)

configs/model/patz.yaml + configs/training/default.yaml + configs/data/hiv_tb.yaml · YAML editable from UI

💻

Hardware

NVIDIA RTX 3050 6GB VRAM · Windows 11 · 16GB RAM · PyCharm IDE · conda environment

Port Map