MindCare AI – HR Copilot for Decision Minds

Predict attrition, personalize growth, and answer HR queries — with generative AI utilizing Gemini 2.5 Pro and alternatively locally with Ollama

MindCare AI is an offline-first HR analytics platform that combines predictive modeling with local Large Language Models to help organizations retain talent, identify growth opportunities, and streamline HR operations without compromising data privacy.

https://huggingface.co/spaces/ganesund/Mindcare_AI

🎯 Key Features

• 🚨 Attrition Risk Radar: Predict attrition with explainable insights
• 📈 Sentiment & Theme Mining: Analyze feedback with local NLP
• 🎓 Career Pathing & Upskilling: AI-powered development recommendations
• 🤖 HR Policy Copilot: RAG-powered chatbot for instant policy Q&A
• 🔒 Privacy-First: 100% local processing, no external API calls

---

🏗️ Architecture

┌────────────────────────────────────────┐
│           HR Data Sources (local)      │
│  • Pulse surveys (CSV)                 │
│  • HRIS exports (CSV/Parquet)          │
│  • PTO & timesheets                    │
│  • L\&D catalog (CSV/Docs)              │
│  • HR policy PDFs/Docs                 │
└────────────────────────────────────────┘
│  (local file share)
▼
┌──────────────────────────────────────────────────────────────────────────┐
│                   Data & Feature Layer (offline)                         │
│  • Ingestion: Python + DuckDB / SQLite                                   │
│  • Cleansing & Anonymization: PII masking, hashing                       │
│  • Features: survey\_sentiment, workload\_ratio, skill\_gap\_score,          │
│              manager\_1on1\_cadence, tenure\_weeks, internal\_moves          │
│  • Embeddings store for RAG: local (FAISS/Chroma on disk)                │
└──────────────────────────────────────────────────────────────────────────┘
│
▼
┌──────────────────────────┐     ┌──────────────────────────────────────────┐
│  Ollama (LLMs local)     │     │   Classical Models (local, Python)      │
│  • mistral / llama3      │     │  • Sentiment: VADER/TextBlob/spaCy      │
│  • codellama for prompts │     │  • Attrition: XGBoost/LogReg (sklearn)  │
│  • RAG over HR docs      │     │  • Recommender: rules + similarity      │
└──────────────────────────┘     └──────────────────────────────────────────┘
│                           │
└──────────────┬────────────┘
▼
┌───────────────────────────────────┐
│     Streamlit/Gradio UI (local)   │
│  • Risk dashboard & drilldowns     │
│  • HR Copilot chat (RAG+LLM)       │
│  • Career paths & export (CSV/JSON)│
└───────────────────────────────────┘

---

🚀 Quick Start

Prerequisites

• Python 3.10+
• Google AI Studio's Gemini 2.5 Pro
• Ollama installed locally
• 8GB+ RAM (recommended), 10GB+ free disk space

Installation

1. Clone the repository

git clone https://github.com/yourusername/mindcare-ai.git
cd mindcare-ai

2. Set up Python environment

python -m venv .venv
# macOS/Linux
source .venv/bin/activate
# Windows (PowerShell)
.\.venv\Scripts\Activate.ps1

pip install -r requirements.txt

3. Install and verify Ollama

# Follow install steps at https://ollama.ai
ollama pull mistral
ollama pull llama3

# Quick check
ollama run mistral "Hello, test message"

4. Prepare sample data (optional)

python scripts/generate_sample_data.py

5. Launch the app

streamlit run app/ui_app.py

6. Open the dashboard at http://localhost:8501

---

📁 Project Structure

mindcare/
├── app/
│   ├── ui_app.py           # Main Streamlit application
│   ├── rag.py              # RAG pipeline for HR Copilot
│   ├── attrition.py        # Attrition prediction models
│   ├── features.py         # Feature engineering utilities
│   ├── recommender.py      # Career pathing recommendations
│   └── storage.py          # Data management utilities
├── data/
│   ├── employees.csv       # Employee master data
│   ├── surveys.csv         # Pulse survey responses
│   ├── timesheets.csv      # Work hours tracking
│   ├── skills.csv          # Employee skills matrix
│   └── policies/           # HR policy documents
├── models/
│   ├── vector_index/       # FAISS/Chroma embeddings
│   └── attrition_lr.joblib # Trained attrition model
├── scripts/
│   ├── generate_sample_data.py
│   └── data_pipeline.py
├── tests/
├── requirements.txt
└── README.md

---

🎮 Usage Guide

1) Risk Radar Dashboard

• Org-wide attrition risk heatmap
• Drilldowns by practice/team/individual
• Top risk drivers (explainable AI)
• Export as CSV

2) Career Pathing

• Personalized development plans
• Course/mentor recommendations
• Project rotation suggestions
• Track skill progression

3) HR Policy Copilot

• Natural-language Q&A over policy docs
• Source citations with RAG
• Handles complex scenarios
• Maintains conversation context

4) Sample Queries

Risk Analysis

• “Show me teams with highest attrition risk”
• “What are the top drivers for the Cloud practice?”

Career Development

• “Recommend growth plan for employee E_1042”
• “What skills are most in demand for Data Engineers?”

Policy Questions

• “What is our parental leave policy?”
• “How many vacation days do I get?”
• “What’s the process for requesting sabbatical?”

---

🔧 Configuration

Environment Variables

Create a .env file in the repo root:

# Ollama Configuration
OLLAMA_HOST=localhost
OLLAMA_PORT=11434
OLLAMA_MODEL=mistral

# Data Configuration
DATA_PATH=./data
VECTOR_INDEX_PATH=./models/vector_index

# Privacy Settings
ENABLE_PII_MASKING=true
ANONYMIZATION_LEVEL=high

Model Configuration

Edit config/models.yaml:

attrition:
  model_type: "logistic_regression"
  features:
    - "tenure_weeks"
    - "survey_sentiment"
    - "workload_ratio"
    - "manager_1on1_cadence"
  threshold: 0.3

rag:
  chunk_size: 512
  chunk_overlap: 50
  top_k: 4
  embedding_model: "sentence-transformers/all-MiniLM-L6-v2"

---

📊 Sample Outputs

Attrition Risk Export (attrition_risk_by_team.csv)

team	avg_risk	top_driver_1	top_driver_2	top_driver_3
Cloud	0.31	workload_ratio	low_1on1	low_sentiment
Data	0.27	low_growth	workload_ratio	pto_spike
AI	0.22	low_recognition	low_1on1	tenure_transition

Career Plan JSON (career_plan_E_1042.json)

{
  "employee_id": "E_1042",
  "current_role": "Data Engineer",
  "target_role": "Senior Data Engineer",
  "skill_gaps": ["Snowflake perf tuning", "dbt testing"],
  "courses": ["Snowflake Performance Deep Dive", "Advanced dbt: Testing & CI"],
  "mentor": "M_309 (Senior DE, Bengaluru)",
  "project_rotation": "FinTech ETL Revamp (4 weeks)"
}

Note: Sample outputs and benchmarks are illustrative—update with your own measurements.

---

🛡️ Privacy & Security

• Local-only processing: No data leaves your infrastructure
• PII masking: Automatic anonymization of sensitive fields
• Role-based access: Configurable permissions
• Audit logging: Track system interactions
• Encryption: Optional at-rest encryption

---

📈 Performance Benchmarks (demo numbers)

Model	Dataset Size	Training Time	Accuracy	Inference Time
Attrition (LogReg)	10K employees	2.3s	0.76 AUC	15ms
Sentiment (VADER)	50K responses	N/A	0.82 F1	5ms
RAG Pipeline	500 documents	45s indexing	~90% hit@k	<2s

---

🧪 Testing

# Unit tests
pytest tests/unit/

# Integration tests
pytest tests/integration/

# End-to-end tests
pytest tests/e2e/

# Performance tests
pytest tests/performance/ --benchmark

---

🤝 Contributing

We welcome contributions!

1. Fork the repo
2. Create a feature branch: git checkout -b feature/amazing-feature
3. Commit: git commit -m "Add amazing feature"
4. Push: git push origin feature/amazing-feature
5. Open a Pull Request

🗺️ Roadmap

Phase 1 (Current)

• ✅ Basic attrition prediction
• ✅ HR policy RAG system
• ✅ Streamlit dashboard
• ✅ Local deployment

Phase 2

• 🔄 Advanced ML models (XGBoost, Neural Networks)
• 🔄 What-if scenario simulation
• 🔄 HRIS integrations
• 🔄 Mobile-responsive UI

Phase 3

• 📋 Fine-tuned domain-specific LLMs
• 📋 Real-time alerting
• 📋 Advanced analytics dashboard
• 📋 Multi-tenant support

---

🎯 Project Goals

• Privacy-First AI: Enterprise-grade AI without data leakage
• Practical Implementation: Actionable HR insights with accessible tech
• Open Innovation: Contribute to the open-source HR analytics ecosystem
• Local-First Architecture: Prove powerful AI can run fully offline

---

📚 Research & Papers

• Predictive HR Analytics
• Local Language Model Deployment
• Privacy-Preserving ML
• Explainable AI in HR Context

---

👥 Team

• Ganesh Sundaresan
• Shashank A
• Simran Singh
• Asma Khanum

---

Made with ❤️ by the MindCare AI Team