Open-source platform tracking how machine learning reshapes pharmaceutical formulation science.
ComplexForm-AI Hub is an open-source academic platform that curates and visualizes the latest ML/AI advances across five complex formulation domains:
- In Situ Gel (原位凝胶)
- Liposome (脂质体)
- Microsphere (微球)
- Nanocrystal (纳米晶)
- PLGA Design (PLGA 设计)
- Paper Library: 90+ curated papers with ML/AI classification, formulation type extraction, and DOI links. Seed papers are integrated by domain (no separate "local library" section) and marked with a Curated badge; daily-sniffed papers are marked Latest.
- Case Study: Interactive walkthrough of FormulationLAI - a full reproduction of the J. Control. Release 389 (2026) 114418 framework for long-acting injectable formulation development (dataset -> ML prediction -> PBPK/PD -> closed-loop optimization -> MD validation)
- ML Foundations: A three-page learning module for formulation scientists new to machine learning:
- ML Basics (
/ml-basics): What is ML vs. traditional programming, three learning paradigms, formulation data types & feature engineering, train/val/test splits, overfitting, cross-validation, evaluation metrics (R², RMSE, MAE, AUC) — with SVG diagrams, formulas, and scikit-learn code snippets - ML Algorithms (
/ml-algorithms): Decision-tree guided algorithm selector, real collection frequency chart (619 papers), and deep-dive cards for 9 method families (Linear/PLS, Random Forest, XGBoost, SVM, ANN, CNN, Gaussian Process & Bayesian Opt, Genetic Algorithm, Clustering & PCA, Generative Models) — each with intuition, formula, runnable code, and a real paper from the hub's collection - ML Workflow (
/ml-workflow): Seven-step project workflow (Define → Collect → Clean → Feature Eng. → Model → Validate → Deploy), QbD/DoE integration diagram, model validation strategies, GxP/CSV compliance essentials, and a curated learning path with recommended books, tools, and paper-reading guide
- ML Basics (
- Daily Sniffer: Automated SCI paper discovery via OpenAlex + PubMed APIs
- LLM Summarization: Two-step prompt chain for structured academic summaries
- Interactive Dashboard: Domain distribution, publication trends, AI method taxonomy
- Knowledge Graph: Force-directed graph showing domain-method relationships
- Bilingual UI: English (default) / Simplified Chinese, switchable at any time
# Backend: parse local PDFs and build database
python src/parser.py
python src/build_taxonomy.py
python src/export_frontend.py
# Frontend: dev server
cd frontend
npm install --legacy-peer-deps
npm run devSee DEPLOY.md for complete deployment instructions.
| Layer | Technology |
|---|---|
| Data sources | OpenAlex API, PubMed E-utilities |
| Backend | Python 3.12, PyMuPDF, SQLite, JSONL |
| LLM | OpenAI-compatible API (DeepSeek / OpenAI / Ark / NVIDIA NIM) |
| Frontend | React 18, Vite 5, TailwindCSS 3, Recharts |
| CI/CD | GitHub Actions |
| Hosting | Hugging Face Spaces (Docker + nginx) |
complexform-ai-hub/
├── src/ # Python backend
│ ├── config.py # Central configuration
│ ├── parser.py # PDF parser + rule-based extraction
│ ├── sniffer.py # OpenAlex/PubMed paper sniffer
│ ├── summarizer.py # LLM two-step prompt chain
│ ├── build_taxonomy.py # Aggregation statistics
│ └── export_frontend.py # Frontend JSON exporter
├── frontend/ # React frontend
│ ├── src/pages/ # 8 pages: Overview, Library, Taxonomy, Graph, CaseStudy, MLBasics, MLAlgorithms, MLWorkflow
│ ├── src/components/ml/ # Shared ML Foundations UI components (CodeBlock, Formula, CaseCard, etc.)
│ ├── public/case-study/ # Standalone FormulationLAI interactive walkthrough
│ ├── Dockerfile # HF Spaces Docker
│ └── nginx.conf # nginx config
├── data/ # Database (JSONL + SQLite + JSON)
├── .github/workflows/ # CI/CD: sniff.yml + deploy.yml
├── STRATEGY.md # Architecture design document
├── DEPLOY.md # Deployment manual
└── .env.example # Environment variable template
MIT