Finalized.

README.md

@@ -1,24 +1,24 @@
 ---
-title: RAGnosis
-emoji: 👁
-colorFrom: red
-colorTo: indigo
-sdk: gradio
-sdk_version: 5.35.0
-app_file: app.py
+title: "RAGnosis"
+emoji: "🧠"
+colorFrom: "red"
+colorTo: "indigo"
+sdk: "gradio"
+sdk_version: "5.35.0"
+app_file: "app.py"
 pinned: false
-license: mit
-short_description: Clinical Query Answering with RAG + MIMIC-IV Notes.
+license: "mit"
+short_description: "Clinical Query Answering with Retrieval-Augmented Generation (RAG) and MIMIC-IV Notes."
 ---
 
-# 🩺 RAGnosis – Clinical Reasoning via Retrieval-Augmented Generation
+# 🧠 RAGnosis – Clinical Reasoning via Retrieval-Augmented Generation
 
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
 [![Python](https://img.shields.io/badge/Python-3.10+-blue.svg)](https://www.python.org/)
 [![Hugging Face](https://img.shields.io/badge/HuggingFace-RAGnosis-blue?logo=huggingface)](https://huggingface.co/spaces/asadsandhu/RAGnosis)
 [![GitHub Repo](https://img.shields.io/badge/GitHub-asadsandhu/RAG--Diagnostic--Assistant-black?logo=github)](https://github.com/asadsandhu/RAG-Diagnostic-Assistant)
 
-> ⚕️ A fully offline-capable, Gradio-powered RAG assistant trained on **annotated clinical notes** from the [MIMIC-IV-Ext-DiReCT](https://github.com/asadsandhu/RAG-Diagnostic-Assistant/blob/main/mimic-iv-ext-direct-1.0.0.zip) dataset to perform explainable diagnostic reasoning.
+> ⚕️ A CPU-ready, Gradio-powered RAG assistant for explainable **clinical diagnosis** using annotated notes from the [MIMIC-IV-Ext-DiReCT](https://github.com/asadsandhu/RAG-Diagnostic-Assistant/blob/main/mimic-iv-ext-direct-1.0.0.zip) dataset.
 
 ---
 
@@ -37,21 +37,21 @@ Try it live on **Hugging Face Spaces** 👉
 
 | Layer        | Details                                                                 |
 |--------------|-------------------------------------------------------------------------|
-| 🧠 Model      | [`Nous-Hermes-2-Mistral-7B-DPO`](https://huggingface.co/NousResearch/Nous-Hermes-2-Mistral-7B-DPO) |
+| 🧠 Model      | [`BioMistral/BioMistral-7B`](https://huggingface.co/BioMistral/BioMistral-7B) |
 | 🏥 Dataset    | [`MIMIC-IV-Ext-DiReCT`](https://github.com/asadsandhu/RAG-Diagnostic-Assistant/blob/main/mimic-iv-ext-direct-1.0.0.zip) |
 | 🔍 Retriever  | FAISS + SentenceTransformers (`all-MiniLM-L6-v2`)                      |
 | 💻 Frontend   | Gradio (Hugging Face Spaces)                                            |
-| 🧠 Backend    | PyTorch + Transformers + BitsAndBytes                                   |
+| 🧠 Backend    | PyTorch + Transformers (no quantization)                               |
 
 ---
 
 ## 🚀 Features
 
-- 🔎 Top-k document retrieval from real annotated clinical notes
-- 📋 Reasoning based on structured diagnostic chains
-- 🧠 GPT-style generation from LLM (Mistral 7B) without internet dependency
-- 🧾 Clean Gradio interface for natural medical queries
-- 🧠 Answers explained like a clinical reasoning expert
+- 🔍 Top-k retrieval from real clinical notes and diagnostic pathways  
+- 📋 Structured reasoning with evidence from retrieved facts  
+- 🧠 Generation powered by domain-specific BioMistral-7B LLM  
+- 💬 Natural question answering with clear clinical explanations  
+- ⚙️ Hugging Face Spaces-friendly: runs on CPU within 16GB RAM  
 
 ---
 
@@ -59,25 +59,24 @@ Try it live on **Hugging Face Spaces** 👉
 
 > *Patient presents with fatigue, orthopnea, and lower extremity edema.*
 
-💬 **Model response:**
-> Based on the patient's symptoms and context, the most likely diagnosis is **congestive heart failure (CHF)**...
+💬 **Model response:**  
+> Based on the patient's symptoms and retrieved clinical facts, the most likely diagnosis is **congestive heart failure (CHF)**...
 
 ---
 
 ## 🛠 How It Works
 
-### ✅ Step 1: Preprocessing
-- Extract chains from `samples/` and `diagnostic_kg/`
-- Build retrievable clinical observations + diagnoses
+### ✅ Step 1: Retrieval (FAISS)
+- Sentence embeddings generated using `all-MiniLM-L6-v2`  
+- Indexed with FAISS (`faiss_index.bin`)  
+- Source corpus: `retrieval_corpus.csv`
 
-### ✅ Step 2: Retrieval (FAISS)
-- Embed notes using `MiniLM-L6-v2`
-- Save as FAISS index → [`faiss_index.bin`](https://github.com/asadsandhu/RAG-Diagnostic-Assistant/blob/main/faiss_index.bin)  
-- Paired with → [`retrieval_corpus.csv`](https://github.com/asadsandhu/RAG-Diagnostic-Assistant/blob/main/retrieval_corpus.csv)
+### ✅ Step 2: Prompt Construction
+- Query + top-5 chunks formatted into a clinical instruction prompt
 
-### ✅ Step 3: Generation
-- Format prompt in `[INST]` syntax
-- Generate diagnosis using `Nous-Hermes-2-Mistral-7B-DPO`
+### ✅ Step 3: Generation (LLM)
+- Prompt fed to `BioMistral/BioMistral-7B`  
+- Diagnosis + explanation generated using `generate()` (no GPU needed)
 
 ---
 
@@ -133,11 +132,12 @@ This project is under the [MIT License](LICENSE).
 
 ## 🙏 Acknowledgments
 
-* MIMIC-IV-Ext-DiReCT: Annotated diagnostic data
-* Hugging Face Transformers + Gradio
+* MIMIC-IV-Ext-DiReCT: Annotated diagnostic corpus
+* Hugging Face Transformers & SentenceTransformers
 * Facebook Research – FAISS
-* Nous Research – Instruction-tuned Mistral model
+* Gradio for UI
+* BioMistral for domain-aligned LLM
 
 ---
 
-> ⚠️ *Disclaimer: This project is for research/demo use only. Not intended for clinical decision-making.*
+> ⚠️ *Disclaimer: This project is for academic demonstration only. It is not approved for clinical use.*

requirements.txt

@@ -4,6 +4,4 @@ faiss-cpu
 torch
 gradio
 accelerate
-sentencepiece
-bitsandbytes
-blobfile
+sentencepiece