The NaS Knowledge Model is a modular Retrieval-Augmented Generation (RAG) system designed for large-scale ingestion, embedding, and fine-tuning of biomedical literature. Built as a research and infrastructure platform, it enables scalable AI training on open-access biomedical content, with autonomous monthly ingestion, fine-tuning, and deployment.
- FastAPI-powered RAG API with top-k semantic retrieval and context-aware generation
- Automated ingestion from PubMed Central using E-Utilities
- Clean chunking and storage by year and month (
data/clean/YYYY/MM) - Vector indexing using FAISS and SentenceTransformers
- Monthly LoRA-based fine-tuning of TinyLlama-1.1B on Apple Silicon
- Automated S3 uploads for all training batches and model adapters
knowledge-model/
├── adapters/ # LoRA adapters per training batch
├── data/
│ ├── clean/YYYY/MM/ # Chunked, cleaned article text
│ └── science_articles/ # JSONL training files (YYYY-MM.jsonl)
├── embeddings/ # Faiss index builder and retriever
├── ingestion/ # PDF parsing, PubMed fetching, S3 upload
├── processing/ # Text cleaning and chunking pipeline
├── training/ # Fine-tuning with PEFT/LoRA
├── api/ # FastAPI RAG endpoint
├── tests/ # Unit tests for embeddings and backend
└── README.md
- Accepts user questions and performs semantic retrieval from embedded biomedical chunks
- Uses LoRA-fine-tuned TinyLlama-1.1B as the response generator
- Auto-trims context for token limits, returns answer with cited sources
- Hosted locally via FastAPI or deployable to cloud
- Adapters are trained monthly on newly ingested biomedical data
- All chunked data is stored in
data/clean/YYYY/MM - Train files written to
data/science_articles/YYYY-MM.jsonl - LoRA adapters are saved and versioned per batch in
adapters/ - All artifacts are uploaded to AWS S3 using the integrated upload module
- Python 3.12
- HuggingFace Transformers + PEFT
- FAISS (vector store)
- FastAPI (API layer)
- PyMuPDF (PDF parsing)
- PubMed E-Utilities (article ingestion)
- AWS S3 (storage backend)
MIT © 2025 NaS Research