Upload README.md with huggingface_hub

README.md

@@ -1,26 +1,37 @@
 ---
-title: BioMed Clinical Intelligence & Oncology RAG Engine
-emoji: 🧬
+title: RAG Engine
+emoji: 🧠
 colorFrom: blue
-colorTo: green
-sdk: static
+colorTo: indigo
+sdk: streamlit
+sdk_version: 1.36.0
+app_file: app.py
 pinned: true
-tags:
-  - rag
-  - biomedicine
-  - oncology
-  - clinical-trials
-  - hybrid-search
 ---
 
-# 🧬 BioMed Clinical Intelligence & Oncology Literature RAG Engine
+# RAG Engine 🧠
 
-Interactive clinical intelligence engine delivering hybrid biomedical semantic search over peer-reviewed oncology literature (PubMed/PMC) and ClinicalTrials.gov registries.
+A modular, educational **Retrieval-Augmented Generation (RAG)** system built to demonstrate core RAG principles:
+1. **Document Ingestion & Chunking** (sliding window with overlap)
+2. **Dense Vector Embeddings** (cosine similarity search)
+3. **Sparse BM25 Keyword Search** (lexical matching for exact terms)
+4. **Hybrid Retrieval with Reciprocal Rank Fusion (RRF)**
+5. **Context-Augmented Generation** with source citations
 
-## 🎯 Architecture Highlights
-* **Hybrid Semantic Retrieval:** Dense vector cosine similarity blended with BM25 lexical keyword matching.
-* **Citation-Grounded Synthesis:** Strict evidence attribution with direct PubMed PMIDs and ClinicalTrials.gov NCT identifiers.
-* **Biomedical Safety Guardrails:** Automated intent classification that rejects direct medical diagnosis prompts while preserving academic literature inquiry.
+### 💻 Live Interactive Demo
+Use the tabs on the left to:
+- **💬 Ask Question**: Query the knowledge base with real-time source grounding and faithfulness scoring.
+- **🔬 Pipeline Inspector**: Directly observe the raw output of Vector Search vs. BM25 Search vs. RRF Rank Fusion.
+- **📚 Knowledge Base**: Inspect loaded documents or upload custom text/PDF files to index them immediately.
 
-Designed & deployed by **[Jayan Gupta](https://huggingface.co/Jayan12321)** (Kaggle: [jayangupta00](https://www.kaggle.com/jayangupta00)).
-Complementary Kaggle Dataset: [Biomedical Oncology Literature & Clinical Trials](https://www.kaggle.com/datasets/jayangupta00/biomedical-oncology-literature).
+### 📐 Architecture
+- **Chunking**: ~400-word passages with 50-word overlap to balance semantic focus and boundary continuity.
+- **Embeddings**: Deterministic token hash embeddings by default (100% offline, zero-dependency), compatible with HuggingFace SentenceTransformers.
+- **Hybrid Fusion**: Merges dense and sparse ranks via Reciprocal Rank Fusion:
+  RRF(d) = \sum_{m} \frac{w_m}{60 + \text{rank}_m(d)}
+- **Generation**: Formats grounded context into system prompts. Provides extractive summaries on CPU or full LLM generation when an API key is provided.
+
+### 🛠️ Transparency & Code
+- **Author**: Jayan Gupta
+- **Source Code & Unit Tests**: [GitHub Repository](https://github.com/JayanGupta/RAG-Engine)
+- **Design Philosophy**: Simple, clean Python without heavy abstraction layers or type boilerplate.