Update README.md

README.md

@@ -12,4 +12,75 @@ short_description: 'AI stock recommendation engine. '
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+---
+title: SectorSync AI
+emoji: 📈
+colorFrom: green
+colorTo: blue
+sdk: gradio
+sdk_version: 4.41.0
+app_file: app.py
+pinned: false
+license: mit
+---
+
+# 📈 SectorSync AI
+
+**Cut Through the Market Noise — Find Your Next Winning Stock in Seconds.**
+
+SectorSync AI is an intelligent, B2B investment matching engine. It leverages a fully synthetic dataset of 12,000 global companies, powered by FAISS Vector Search and state-of-the-art Generative AI, to instantly connect investors with high-growth companies that perfectly match their investment thesis.
+
+## 🧠 The AI Architecture
+
+Our recommendation pipeline utilizes a three-tier AI architecture to ensure speed, accuracy, and logic:
+1. **Embedding Layer:** We use `SentenceTransformer` (`paraphrase-MiniLM-L3-v2`) to map 12,000 company profiles into a 384-dimensional semantic space.
+2. **Vector Retrieval:** We utilize `FAISS` (Facebook AI Similarity Search) to perform lightning-fast L2-distance searches, instantly retrieving the closest matching companies to the user's enriched query.
+3. **Generative Sales Pitch:** We integrated a state-of-the-art Causal LLM (**Qwen2.5-0.5B-Instruct**) to act as a Wall Street analyst. Using ChatML templates and strict PyTorch determinism, the model generates a highly aggressive, logical, and custom 1-sentence sales pitch for every retrieved company.
+
+---
+
+## 📊 Data Science Benchmarks & Exploratory Data Analysis
+
+Before deploying this application, the underlying embedding space was rigorously tested and validated in our Exploratory Data Analysis (EDA) phase to ensure the AI separates true signal from market noise.
+
+### 1. Supervised ML Leaderboard (Proving Semantic Understanding)
+
+*(Upload your ML Leaderboard picture to Hugging Face and put the filename inside the parentheses below)*
+![ADD PICTURE HERE - Supervised ML Leaderboard](your_ml_leaderboard_image.png)
 
+* **What it means:** This plot proves that our AI embeddings successfully capture deep semantic meaning. Because the **k-Nearest Neighbors (k-NN)** model achieved **74.2% accuracy** classifying companies into 6 major parent domains, it demonstrates that the AI truly understands the underlying business models rather than just matching superficial keywords.
+
+### 2. Enhanced AI Recommendation Accuracy by Sector (Proving Precision)
+
+*(Upload your Sector Accuracy picture to Hugging Face and put the filename inside the parentheses below)*
+![ADD PICTURE HERE - Enhanced AI Recommendation Accuracy by Sector](your_sector_accuracy_image.png)
+
+* **What it means:** This proves the absolute precision of our retrieval system. By simulating 250 real-world investor queries using "Query Enrichment", the engine achieved **100% accuracy** in returning companies from the exact target sector, guaranteeing that users are never recommended irrelevant background noise.
+
+### 3. AI Match Confidence vs. Random Noise (Proving Signal Separation)
+
+*(Upload your Histogram picture to Hugging Face and put the filename inside the parentheses below)*
+![ADD PICTURE HERE - AI Match Confidence vs. Random Noise](your_histogram_image.png)
+
+* **What it means:** This plot proves mathematical signal separation. It demonstrates that when our AI finds a matching company, the semantic similarity score is statistically distinct from the background noise of random companies. This guarantees our recommendation engine is detecting a true "investment signal" rather than just guessing.
+
+### 4. Unsupervised AI Structure: K-Means Clustering (Proving Data Quality)
+
+*(Upload your PCA Scatter Plot picture to Hugging Face and put the filename inside the parentheses below)*
+![ADD PICTURE HERE - K-Means Clustering in PCA Space](your_pca_cluster_image.png)
+
+* **What it means:** This proves that even without being given any labels (unsupervised learning), the AI naturally understands the global economy. By compressing 384-dimensional embeddings into 2D space, the K-Means algorithm independently discovered distinct economic clusters, validating the high quality, diversity, and richness of our synthetic 12,000-company dataset.
+
+---
+
+## 🚀 How to Use the App
+
+1. **Enter a Thesis:** Type an investment thesis (e.g., *"autonomous robotics for supply chains"*).
+2. **Filter by Sector (Optional):** Narrow down your search using the sector dropdown menu.
+3. **Search:** Click "Find Investment Matches".
+4. **Results:** The AI will return the top companies, their similarity scores, and a dynamically generated GenAI sales pitch explaining exactly why they fit your thesis.
+
+*(Note: The very first query may take ~10 seconds as the Hugging Face ZeroGPU lazy-loads the GenAI model into memory. All subsequent queries are instant!)*
+
+---
+*Developed as a Final Project for University Data Science.*