Curated resources, architectures, prompts, and benchmark datasets for applying LLMs and Retrieval-Augmented Generation (RAG) in biomedicine and drug discovery.
-
Updated
Aug 8, 2026 - Python
Curated resources, architectures, prompts, and benchmark datasets for applying LLMs and Retrieval-Augmented Generation (RAG) in biomedicine and drug discovery.
Automated pharmacovigilance literature validation for ultra‑rare neurological diseases. Enter a drug name and an adverse event. The app searches PubMed with a multi‑tier cascade, extracts the most relevant evidence, and returns a verdict — Present or Absent — backed by a fine‑tuned PubMedBERT model.
AI-powered conversational medical assistant with RAG over MedLinePlus & MedQuAD datasets, Neo4j knowledge graph, Qdrant vector database, LangGraph agent orchestration, and voice I/O (Whisper STT + Edge TTS).
Multimodal AI-powered plant disease diagnosis system integrating ResNet-18, Quantum Neural Networks (PennyLane), PubMedBERT, Gemini 2.5 Flash, Grad-CAM, and probabilistic ensemble fusion for explainable disease classification using computer vision, NLP, and quantum-enhanced deep learning.
Transformer-based topic modeling (BERTopic) of six spine journals, 1978–2026 — 40,401 abstracts, 37 topics, with temporal trends and VOSviewer comparison.
Fine-tuned PubMedBERT bi-encoder on PubMedQA achieving NDCG@10=0.974. BM25 from scratch in pure NumPy. Self-improving multi-agent critic loop. Model live on HuggingFace Hub.
Fine-tuned PubMedBERT for medical abstract sentence classification. 89% accuracy on PubMed 200k RCT
Classifying NHS patient questions by disease — TF-IDF+kNN baseline vs. fine-tuned BERT / BioBERT / PubMedBERT (UCL MSc, Advanced ML for Healthcare).
Local multi-agent pipeline (LLaMa-3.1 via MLX + PubMedBERT RAG over Chroma) that fact-checks drug questions against authoritative medical sources and returns a cited answer with per-claim verdicts. Ships with a FastAPI + single-page web UI and a Typer CLI
Agentic Corrective RAG over 484 ClinicalTrials.gov oncology protocols. LangGraph CRAG with LLM-as-judge grading + cross-encoder reranker for two-stage retrieval. Three independent refusal gates. PubMedBERT, FAISS, Groq Llama 3.3 70B. Live demo on Streamlit.
An AI agent system that reasons about a drug program end to end: probability-of-success anchored to real trial outcomes (PubMedBERT + LightGBM on HINT), phase-gated rNPV, and an 8-pillar clinical/regulatory scorecard — forming a falsifiable, cited view. FastAPI + Next.js, three runtime agents.
Add a description, image, and links to the pubmedbert topic page so that developers can more easily learn about it.
To associate your repository with the pubmedbert topic, visit your repo's landing page and select "manage topics."