Add HuggingFace Spaces metadata to README.md

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

README.md

@@ -1,200 +1,40 @@
-# Verification-Driven Hallucination Firewall (VDHF)
-
-A modular Python system that verifies RAG (Retrieval-Augmented Generation) outputs before delivering them to users, preventing AI hallucinations.
-
----
-
-## 📋 Prerequisites
-
-### Required
-1. **Python 3.9+** - Download from https://www.python.org/downloads/
-   - ⚠️ During installation, check **"Add Python to PATH"**
-
-### Optional (for full LLM features)
-2. **OpenAI API Key** - Get from https://platform.openai.com/api-keys
-   - Without this, the system uses mock generation for testing
-
----
-
-## 🚀 Quick Start
-
-### Step 1: Install Dependencies
-```powershell
-cd "c:\Users\HP\Desktop\Hallucination Firewall"
-pip install -r requirements.txt
-```
-
-### Step 2: (Optional) Configure API Key
-```powershell
-copy .env.example .env
-# Edit .env and add your OpenAI API key
-```
-
-### Step 3: Run the System
-```powershell
-python main.py
-```
-
----
-
-## 📂 Project Structure
-
-```
-Hallucination Firewall/
-│
-├── config.py              # Configuration (thresholds, models)
-├── ingestion.py           # Document loading (PDF, TXT, DOCX)
-├── embeddings.py          # Sentence-BERT embeddings + ChromaDB
-├── retriever.py           # Semantic search for evidence
-├── generator.py           # LLM response generation
-├── claim_extractor.py     # Extract factual claims from text
-├── verifier.py            # Verify claims using similarity + NLI
-├── firewall.py            # Decision engine (pass/block)
-├── prompt_refiner.py      # Regenerate safer responses
-├── main.py                # Main pipeline + interactive CLI
-│
-├── requirements.txt       # Python dependencies
-├── .env.example           # API key template
-│
-├── sample_docs/
-│   └── sample.txt         # Sample test documents
-│
-└── tests/
-    └── test_pipeline.py   # Unit tests
-```
-
----
-
-## 🔧 How It Works
-
-```
-┌─────────────┐    ┌──────────────┐    ┌─────────────────┐
-│ User Query  │───▶│   Retrieve   │───▶│ Generate Answer │
-└─────────────┘    │   Evidence   │    └────────┬────────┘
-                   └──────────────┘             │
-                                                ▼
-┌─────────────┐    ┌──────────────┐    ┌─────────────────┐
-│   Output    │◀───│   Firewall   │◀───│ Extract Claims  │
-│  (Verified) │    │   Decision   │    │   & Verify      │
-└─────────────┘    └──────────────┘    └─────────────────┘
-                          │
-                          │ If fails threshold
-                          ▼
-                   ┌──────────────┐
-                   │   Refine &   │
-                   │  Regenerate  │
-                   └──────────────┘
-```
-
-### Pipeline Steps:
-1. **Ingest Documents** → Load PDFs/TXT files into vector database
-2. **Retrieve Evidence** → Find relevant chunks for user query
-3. **Generate Response** → LLM creates initial answer
-4. **Extract Claims** → Break response into atomic factual statements
-5. **Verify Claims** → Check each claim against evidence
-6. **Firewall Decision** → Pass if ≥80% claims verified
-7. **Regenerate** → If failed, refine prompt and try again
-
 ---
-
-## 💻 Interactive Commands
-
-When running `main.py`, use these commands:
-
-| Command | Description |
-|---------|-------------|
-| `/ingest <path>` | Load documents from file or folder |
-| `/clear` | Clear all documents from memory |
-| `/count` | Show number of document chunks |
-| `/quit` | Exit the program |
-
-**Example session:**
-```
-You: /ingest sample_docs
-Ingested sample_docs: 5 chunks total
-
-You: When was Python released?
-[Processing...]
-✓ VERIFIED - Support Ratio: 100%
-Response: Python was first released in 1991 by Guido van Rossum.
-```
-
----
-
-## ⚙️ Configuration
-
-Edit `config.py` to customize:
-
-| Setting | Default | Description |
-|---------|---------|-------------|
-| `SIMILARITY_THRESHOLD` | 0.75 | Minimum similarity for claim-evidence match |
-| `FIREWALL_THRESHOLD` | 0.8 | Minimum % of claims that must be verified |
-| `TOP_K_RETRIEVAL` | 7 | Number of evidence chunks to retrieve |
-| `CHUNK_SIZE` | 1000 | Characters per document chunk |
-| `LLM_MODEL` | gpt-3.5-turbo | OpenAI model to use |
-
----
-
-### Step 4: Run Interactive Querying
-To interactively query the system and see verification results:
-```powershell
-python run.py --demo
-```
-
----
-
-## 🧪 Running Tests
-
-
-```powershell
-python -m pytest tests/test_pipeline.py -v
-```
-
+title: Hallucination Firewall
+emoji: 🛡️
+colorFrom: blue
+colorTo: indigo
+sdk: docker
+pinned: false
 ---
 
-## 📝 Example Output
-
-```
-============================================================
-VDHF Pipeline Result
-============================================================
-Status: ✓ VERIFIED
-Support Ratio: 100.00%
-Claims: 2/2 supported
-Regeneration Attempts: 0
-============================================================
-Response:
-Python was released in 1991. It was created by Guido van Rossum.
-============================================================
-```
+# Verification-Driven Hallucination Firewall (VDHF)
 
----
+A modular Python system that verifies RAG (Retrieval-Augmented Generation) outputs before delivering them to users, preventing AI hallucinations.
 
-## 🔑 What You Need to Provide
+Upload documents (TXT, PDF, DOCX, Excel, CSV), ask questions, and get verified answers with every claim checked against your content.
 
-| Item | Required? | How to Get |
-|------|-----------|------------|
-| Python 3.9+ | ✅ Yes | https://python.org/downloads |
-| Documents to verify against | ✅ Yes | Your PDFs, TXT files |
-| OpenAI API Key | ❌ Optional | https://platform.openai.com/api-keys |
+## How It Works
 
----
+1. **Upload Documents** - Upload any document to the system
+2. **Ask Questions** - Query your uploaded content
+3. **Claim Extraction** - Every factual claim in the response is identified
+4. **Verification** - Each claim is checked against your uploaded data
+5. **Firewall Decision** - Response is marked as Verified, Partially Verified, or Hallucinated
+6. **Regeneration** - If needed, a safer response is generated
 
-## 📚 Module Details
+## Features
 
-### claim_extractor.py
-Extracts atomic factual claims from LLM responses:
-- Filters out opinions ("I think...")
-- Splits compound sentences ("X and Y" → "X", "Y")
-- Identifies verifiable statements
+- Excel/CSV direct data analysis (no ML models needed)
+- Student comparison and filter queries
+- Claim verification against uploaded data
+- Hallucination detection for non-existent records
+- Groq LLM-powered analysis for complex questions
+- Beautiful React frontend with tabular response rendering
 
-### verifier.py
-Two-step verification:
-1. **Semantic Similarity**: Cosine similarity ≥ 0.75
-2. **NLI Entailment**: Evidence must logically support claim
+## Tech Stack
 
-### firewall.py
-Decision logic:
-- Calculate `SupportRatio = supported_claims / total_claims`
-- If ratio ≥ 0.8: **PASS** (deliver to user)
-- If ratio < 0.8: **REGENERATE** (refine and retry)
+- **Backend**: FastAPI + Python
+- **Frontend**: React + Vite + Tailwind CSS
+- **ML**: Sentence-BERT, DeBERTa NLI
+- **Vector DB**: ChromaDB
+- **LLM**: Groq API