How vidyutchakrabarti/kenkomirai works
Project type and purpose
• This is a full-stack, modular simulation platform (“COVID Digital Twin & Containment Simulation”) combining:
• A Next.js frontend for interactive dashboards/maps/charts
• A FastAPI backend hosting the agent-based simulation and chain-of-thought logic
• An ML/RL module (using OpenAI Gym and stable-baselines3) for training disease-containment policies
• Infrastructure (Docker Compose & Kubernetes) to orchestrate services
File structure
• Top-level directories:
• assets (static images)
• backend (FastAPI app, API endpoints, simulation code, data ingestion/preprocessing, tests)
• frontend (Next.js app with components, pages, styles)
• infrastructure (docker-compose.yml, k8s manifests, deploy scripts)
• initial datasets (shapefiles, census data)
• ml (training scripts for RL agents)
• Patterns & config files:
• backend/app/api → REST endpoints
• backend/app/simulation → agent-based model (environment, agent, profiles, RL wrapper)
• backend/app/data → ingest & preprocess raw GIS/census data
• ml/training/train_rl_agent.py → custom Gym environment integration
• infrastructure/docker-compose.yml + k8s/ → multi-service orchestration
• frontend uses Leaflet, Chart.js, Tailwind, Next.js routing
README for additional insights
• Confirms a 3-module architecture (Frontend / Backend / ML-RL)
• Highlights external dependencies:
• LLMs for chain-of-thought routines
• OpenAI Gym interface for RL agents (e.g., PPO)
• Visualization libs (Leaflet, Chart.js)
• Describes data flow: agent initialization → simulation → RL policy loop → frontend visualization