effiHR

effiHR is an agentic HR performance intelligence project that combines dataset processing, multi-agent analysis, and a Streamlit dashboard.

The project workflow is:

• Build a unified HR dataset from source files.
• Run a CrewAI pipeline to generate performance bundles, insights, review drafts, and alerts.
• Explore results in an interactive dashboard.

Key Features

• Multi-agent pipeline orchestration from main.py
• Tooling for HR data access, GitHub activity, and Slack notifications in tools.py
• Data unification script in process_hr_data.py
• Dashboard UI in app.py
• JSON artifacts written to output/

Repository Layout

• app.py: Streamlit dashboard
• main.py: Pipeline runner
• config.py: Environment and runtime configuration
• tools.py: CrewAI tools (HRMS, GitHub, Slack, output persistence)
• process_hr_data.py: Builds data/unified_hr_data.json from files in dataset/
• inspect_datasets.py: Quick dataset inspection helper
• dataset/: Source datasets (CSV/XLSX)
• data/: Unified data and optional raw copies
• output/: Pipeline outputs and state files

Prerequisites

• Python 3.11+
• pip

Setup

1. Create and activate a virtual environment:

python3 -m venv .venv
source .venv/bin/activate

2. Install dependencies (no lockfile/requirements file is currently committed):

pip install --upgrade pip
pip install crewai streamlit pandas plotly python-dotenv requests PyGithub openpyxl

3. Create a .env file in the project root:

NVIDIA_API_KEY=your_nvidia_api_key
GITHUB_TOKEN=
GITHUB_ORG=
SLACK_WEBHOOK_URL=

Notes:

• NVIDIA_API_KEY is required to run the main pipeline.
• GITHUB_TOKEN and SLACK_WEBHOOK_URL are optional. If absent, the project falls back to mock GitHub data and skips Slack sends.

Quick Start

1. Optional: inspect source datasets

python inspect_datasets.py

2. Build unified HR data

python process_hr_data.py

3. Run the pipeline

python main.py

4. Launch dashboard

streamlit run app.py

Output Files

The pipeline writes JSON artifacts to output/:

• performance_bundles.json: Employee-level consolidated signals
• insight_reports.json: Scored analysis and trend insights
• review_drafts.json: AI-generated review drafts and rating rationale
• alerts.json: Alert objects for nudges, risks, and quality issues
• pipeline_state.json: Current/last run state metadata
• benchmark_results.json: Benchmark timing snapshot (if produced in your run)

Configuration Notes

• Core settings live in config.py.
• PIPELINE_MAX_EMPLOYEES limits how many employees are processed in one run.
• Performance score weights are validated to sum to 1.0.

Troubleshooting

• If you see NVIDIA_API_KEY is not set, verify your .env file exists and contains NVIDIA_API_KEY.
• If dependencies fail to import, activate your virtual environment and reinstall the packages listed above.
• If output files look stale, rerun python main.py and refresh the dashboard.