GENIUS orchestrates VASP, Quantum ESPRESSO, LAMMPS, SPPARKS, and GROMACS via a 6-stage AI recommendation pipeline — from natural language intent to validated simulation inputs — powered by Smart Knowledge Graphs, Hybrid Search, and Multi-LLM hierarchies.
From knowledge graph parameter retrieval to finite state machine execution. Hover & drag to inspect live interactive engines.
Apache AGE + pgvector graph database with TXC paramjson trees, taxonomy crosswalks, and intent vocabularies across 1,600+ parameters for VASP, QE, LAMMPS, and SPPARKS.
6-stage pipeline: Intent Extraction → Seed Discovery → Dependency Expansion → Value Assignment → Validation → Output Generation. OpenRouter with heuristic model routing and batch concurrency.
VASP INCAR generation (558 tags) with version-gating (6.4/6.5), SOC/hybrid/GW compatibility. QE pw.x inputs (240 params) with k-grid, pseudopotential, and DFT+U validation.
LAMMPS script generation (768 commands, Intel build) with EAM/Tersoff/ReaxFF/NEB support. SPPARKS input generation (101 commands) for Ising, Potts, diffusion, and AM melt-pool simulations.
ONNX embedding model (all-MiniLM-L12-v2) with pgvector similarity search, full-text PostgreSQL indexing, per-program search indexes, and context-aware snippet retrieval.
Containerized simulation execution via Docker (LAMMPS, QE, VASP images) or SSH+Slurm dispatch to HPC clusters. MPI topology, NEB replica partitioning, and automatic job staging.
R-based classical DoE via Docker runtime, surrogate modeling, Bayesian optimization campaigns, and main-effects analysis for automated experiment planning.
Vue 3 + Vite interactive graph visualization, Express API backend, MediaWiki knowledge base with bot-driven ingestion, and taxonomy-driven parameter navigation.
From natural-language intent to validated simulation input files — fully automated parameter discovery, expansion, and assignment.
Generate validated simulation input files from plain-text calculation requests. GENIUS handles intent extraction, parameter discovery, dependency expansion, LLM assignment, and compatibility validation across all supported simulation codes.
# POST /workflow/ — Submit Calculation Intent to FastAPI Backend (port 8001)
curl -X POST "http://127.0.0.1:8001/workflow/" \
-H "Content-Type: application/json" \
-d '{
"calculation_prompt": "Relax CeO2 with Quantum ESPRESSO & compute MD with LAMMPS",
"lm_api": "your_openrouter_api_key",
"gen_model_hierarchy": ["databricks/dbrx-instruct", "meta-llama/llama-3.1-405b-instruct"],
"project_config": { "name": "CeO2_Cloud_Job", "num_retries_permodel": 3 }
}'
# Monitor Real-Time Log Stream (Server-Sent Events)
curl -N "http://127.0.0.1:8001/logs"
# Retrieve Status & Computed Results
curl "http://127.0.0.1:8001/workflow-status/{workflow_id}"
curl "http://127.0.0.1:8001/results/{workflow_id}" | jqStreamline your computational research across VASP, Quantum ESPRESSO, LAMMPS, SPPARKS, and GROMACS with the GENIUS autonomous lab monorepo.