Update README.md

README.md

@@ -1,106 +1,109 @@
-# 🧠 Mental Health Counselor Assistant
-
-A privacy-friendly, intelligent assistant designed to help mental health professionals explore **response suggestions** based on patient input. Built with `XGBoost` for intent classification and `Mistral-7B` for natural language generation, this app helps support clinical conversations with empathy and structure — while keeping you in full control.
-
 ---
-
-## ✅ Problem Statement
-
-Mental health counselors are tasked with responding to a wide range of complex emotions and situations. In high-pressure or uncertain moments, this tool helps:
-
-- Predict the **intent** behind a patient's message (advice-seeking, validation, informational, or inquisitive)
-- Generate a supportive, **AI-assisted counselor response**
-- Log conversations for review and learning
-- Preserve privacy with local inference — built for **HIPAA-aware workflows**
-
+title: Counselor Assistant
+emoji: 🧠
+colorFrom: purple
+colorTo: blue
+sdk: streamlit
+sdk_version: "1.32.2"
+app_file: app.py
+pinned: false
+---
+title: Counselor Assistant
+emoji: 🧠
+colorFrom: purple
+colorTo: blue
+sdk: streamlit
+sdk_version: "1.32.2"
+app_file: app.py
+pinned: false
 ---
 
-## ✨ Features
+# 🧠 Counselor Assistant - AI Support for Mental Health Conversations
 
-- 🔍 **Intent Prediction** — XGBoost classifier trained on annotated mental health dialogue
-- 💬 **Response Generation** — LLM-backed replies using quantized [Mistral-7B](https://mistral.ai/news/announcing-mistral-7b/)
-- 🧠 **Multi-turn Memory** — Maintains up to 6 rounds of conversation
-- ✅ **Export to JSON + CSV** — With timestamps for traceability
-- 🧰 **HIPAA-conscious Design** — No third-party API calls required; all runs locally
+This Streamlit app helps **mental health counselors** explore supportive language and therapeutic approaches for responding to patient concerns. It leverages a **machine learning classifier** (XGBoost) to detect intent and a **quantized Mistral-7B model** to generate helpful replies.
 
 ---
 
-## 🖼️ App Preview
+## ✅ Features
 
-![counselor assistant demo](demo/demo.gif)
+- **ML-Based Intent Prediction**: Uses TF-IDF + XGBoost to detect if the message needs validation, advice, information, or follow-up.
+- **Mistral-7B Counselor Response**: LLM-powered suggestion using your local model (gguf) or Hugging Face-hosted model.
+- **Multi-turn Memory**: Keeps context from recent messages.
+- **Word Count + Guidance**: Shows word count to help user compose inputs.
+- **Conversation Export**: Save to `.csv` or `.json` for recordkeeping.
+- **Interview-Ready Design**: Friendly UI, emoji icons, prompt examples, and reset options.
 
 ---
 
 ## 🚀 How to Run
 
-### 🔧 Local (with Mistral-7B)
+### 🧪 Locally (with your own Mistral model)
 
-> 🧩 You must download a quantized `.gguf` model file from Hugging Face or TheBloke, e.g.:
->
-> `/Users/yourname/models/mistral/mistral-7b-instruct-v0.1.Q4_K_M.gguf`
+1. Clone the repo:
+   ```bash
+   git clone https://github.com/your-username/mental_health_chatbot.git
+   cd mental_health_chatbot
+   ```
 
-```bash
-git clone https://github.com/yourname/mental-health-chatbot.git
-cd mental-health-chatbot
-pip install -r requirements.txt
-streamlit run app_local.py
-```
+2. Place your **gguf model** under `models/`:
+   ```
+   models/mistral-7b-instruct-v0.1.Q4_K_M.gguf
+   ```
+
+3. Create and activate a virtual environment:
+   ```bash
+   pip install -r requirements.txt
+   streamlit run app.py
+   ```
 
 ---
 
-### 🐳 Docker (Recommended)
+### 🌐 Deploy on Hugging Face Spaces
 
-```bash
-docker build -t counselor-assistant .
-docker run -p 8501:8501 counselor-assistant
-```
+- ✅ Use this repo
+- ✅ Upload your model or switch to a hosted one like `google/flan-t5-base`
+- ✅ Add a `Space` on [Hugging Face](https://huggingface.co/spaces)
+
+> You can modify `generate_llm_response()` to use Hugging Face's `pipeline` if GPU is not needed.
 
 ---
 
-## 📁 Project Structure
+## 📁 Folder Structure
 
 ```
-mental-health-chatbot/
+mental_health_chatbot/
 │
-├── app_local.py              # Streamlit app (uses llama-cpp + XGBoost)
+├── app.py
 ├── requirements.txt
 ├── Dockerfile
-├── demo/                     # Optional: screenshots or .gif
-├── dataset/
+├── README.md
+├── LICENSE
+├── data/
 │   └── Kaggle_Mental_Health_Conversations_train.csv
 ├── models/
 │   └── mistral-7b-instruct-v0.1.Q4_K_M.gguf
-├── README.md
-└── LICENSE
+├── utils/
+│   └── helper_functions.py
+└── static/
+    └── images/
+        └── demo.gif
 ```
 
 ---
 
-## 📌 Design Decisions
-
-- 🧠 Used `TfidfVectorizer` + `XGBoost` as a **lightweight, interpretable** classifier
-- 🔒 LLM inference handled locally with `llama-cpp-python` — no external data leaks
-- 📦 Docker support to simulate realistic deployment and improve portability
-- 🎯 Optimized for educational and prototyping use in **clinician settings**
-
----
-
-## 📍 Future Improvements
+## 📸 Screenshot / Demo
 
-- Add long-form audio transcription support
-- Integrate feedback loop for model tuning
-- Auto-summary of sessions
-- Hugging Face Space version (hosted with opt-in privacy tradeoffs)
+![demo](static/images/demo.gif)
 
 ---
 
-## 👩‍⚕️ A Note to Reviewers
+## 👩‍💻 Why I Built This
 
-This project was developed for an interview to showcase how **AI and empathy** can work hand-in-hand to support mental health professionals. Thank you for reviewing, and I hope it sparks a great conversation about the intersection of tech and care. 💙
+Mental health professionals often face emotionally complex conversations. This assistant offers **non-binding, thoughtful suggestions** to aid their decision-making — not to replace them, but to **support their professional judgment**.
 
 ---
 
-## 📄 License
+## 📜 License
 
-MIT — free to use and modify, but please be thoughtful in healthcare applications.
+MIT License — see [`LICENSE`](./LICENSE)