Update README.md

README.md

@@ -9,13 +9,16 @@ app_file: app.py
 pinned: true
 tags:
 - mcp-in-action-track-consumer
+- sambanova
+- nebius
+- huggingface
 - depression-detection
 - mental-health
 - longformer
 - mcp
 - agents
 license: mit
-short_description: MCP-enabled depression detection agent
+short_description: Agentic depression detection using Kimi K2/Llama 3
 ---
 
 # 🧠 Early Depression Detection MCP Agent
@@ -26,42 +29,35 @@ short_description: MCP-enabled depression detection agent
 ## 📹 Demo Video
 [INSERT YOUTUBE/LOOM LINK HERE]
 
-## 📱 Social Media Post
-[INSERT LINKEDIN/X POST LINK HERE]
-
 ## 🎯 Project Description
-This MCP-enabled agent detects depression risk from social media text using Mental-Longformer, achieving **F1-score of 0.7668** on eRisk 2025 test data.
+This MCP-enabled agent detects depression risk from social media text using **Mental-Longformer** as a clinical tool and **SambaNova/Nebius** as reasoning engines. It achieves an **F1-score of 0.7668** on eRisk 2025 test data.
 
-**Key Features:**
-- 🔍 4,096-token context window (8x BERT's capacity)
-- 📊 Trained on eRisk 2017-2022 datasets
-- 🤖 Data augmentation with Gemini 2.5 Flash Lite
-- ⚡ Real-time linguistic pattern analysis
-- 🔌 MCP-enabled for agent integration
+### 🤖 Multi-Provider Agent Architecture
+This agent demonstrates **MCP integration** by orchestrating specialized tools with high-performance LLMs:
 
-**Model:** [avtak/erisk-longformer-depression-v1](https://huggingface.co/avtak/erisk-longformer-depression-v1)
+1.  **Tool Layer (The "Eyes"):**
+    *   **Model:** `avtak/erisk-longformer-depression-v1` (My fine-tuned model).
+    *   **Capability:** 4,096-token context window for analyzing long user timelines.
+    *   **Logic:** Implements thesis findings (Echo Chamber effects, Nocturnal patterns).
 
-## 🧪 How It Works
-The agent analyzes long-form text for linguistic markers including:
-- Anhedonia (loss of interest)
-- Self-focused negative language
-- Social withdrawal indicators
-- Hopelessness themes
-- Disrupted sleep/energy patterns
+2.  **Reasoning Layer (The "Brain"):**
+    *   **User Choice:** The agent allows switching between reasoning providers.
+    *   **SambaNova:** Uses `Meta-Llama-3.1-70B-Instruct` for lightning-fast inference.
+    *   **Nebius:** Uses `moonshotai/Kimi-K2-Thinking` for advanced reasoning.
 
-## 🏆 Research Background
-Built on Master's thesis research at **University of Malaya**, this model addresses critical challenges in early depression detection:
-- Handles imbalanced datasets through LLM-powered augmentation
-- Captures long-context dependencies (4096 vs 512 tokens)
-- Rigorous 5-fold cross-validation (mean F1: 0.862, std: 0.009)
-- Validated on held-out eRisk 2025 test set
+## 🧪 Thesis Findings Integrated
+The agent doesn't just output a score; it looks for specific behavioral biomarkers identified in my Master's Thesis:
+*   **"High-Effort, Low-Frequency":** Depressed users write 33% longer posts but post less often.
+*   **"The Homophily Cascade":** Users in the "Moderate" risk zone often engage in "Echo Chambers" (11.7x higher interaction with depressed users).
+*   **"Nocturnal Posting":** High-risk users show distinct activity spikes between 00:00 - 05:00 UTC.
 
-## 👥 Team
-- Hassan Hassanzadeh Aliabadi (@avtak)
+## 🏆 Research Background
+Built on Master's thesis research at **University of Malaya**:
+- **Data Augmentation:** Used Gemini 2.5 Flash Lite to balance the dataset.
+- **Validation:** Rigorous 5-fold cross-validation (mean F1: 0.862).
 
 ## ⚠️ Ethical Considerations
-This is a research tool, **not a medical diagnostic instrument**. Always consult qualified healthcare professionals for mental health concerns.
-
+This is a research tool, **not a medical diagnostic instrument**. 
 **Crisis Resources:**
 - 🆘 Crisis Text Line: Text HOME to 741741 (US)
-- 🌍 International: [befrienders.org](https://befrienders.org)
+- 🌍 International: [befrienders.org](https://befrienders.org)