An open-source workforce intelligence system for SMEs — Prophet labor cost forecasting, PuLP schedule optimization, burnout risk scoring, and CrewAI multi-agent orchestration for workforce analytics.
PRISM is an open-source productivity and labor cost intelligence system. It tracks employee productivity metrics, forecasts labor costs with time-series modeling (now with exogenous data integration), optimizes workforce schedules using linear programming (with US Labor Law Compliance constraints), scores burnout risk to trigger gig worker webhooks, and uses a CrewAI multi-agent layer for conversational orchestration.
Registers employees with role, cost center, contract type, and labor cost metadata. Captures productivity time-series per worker and team. Ingests industry benchmark data for comparative analysis. Persists all records to a local SQLite database.
Input: Employee records · productivity logs · industry benchmark data
Output: Workforce database — registered, time-series structured, benchmark-enriched
Trains Facebook Prophet models on labor cost history to produce seasonal-aware forecasts. Runs PuLP linear programming optimization to minimize scheduling costs while satisfying coverage constraints. Scores burnout risk with scikit-learn classifiers trained on workload, tenure, and overtime patterns. Computes KPIs per employee, team, and cost center.
Input: Workforce database + productivity time-series
Output: Labor cost forecast · optimized schedule · burnout risk scores · productivity KPIs
CrewAI orchestrates six specialist agents across workforce analytics tasks. Generates natural language workforce narratives using Anthropic Claude (or offline rule-based heuristics). Produces benchmark comparison reports by industry sector. Surfaces scheduling and staffing recommendations with explainable justifications.
Input: Forecasts + optimization results + risk scores + benchmark data
Output: Workforce narrative · benchmark report · scheduling recommendations · staffing alerts
graph TD
A["Employee Data Input"] --> B["Productivity Tracker<br/>app/agents/productivity_tracker.py"]
C["Industry Benchmark Data"] --> D["Benchmark Lookup<br/>app/agents/benchmark_lookup.py"]
B --> E[("SQLite Database<br/>prism_workforce.db")]
D --> E
E --> F["Labor Forecaster<br/>app/agents/labor_forecaster.py<br/>Prophet"]
E --> G["Scheduling Optimizer<br/>app/agents/scheduling_optimizer.py<br/>PuLP linear programming"]
E --> H["Burnout Risk<br/>app/agents/burnout_risk.py<br/>scikit-learn"]
F --> I["CrewAI Orchestrator<br/>app/agents/orchestrator.py"]
G --> I
H --> I
D --> I
I --> J["Claude LLM Narrative<br/>/ Offline Heuristics"]
I --> K["FastAPI Backend<br/>app/main.py"]
K --> L["REST API /api"]
L --> M["Glassmorphic Dashboard<br/>frontend/ · Chart.js"]
| Endpoint | Method | Description |
|---|---|---|
/api/employees |
GET / POST |
Employee registry — list or register |
/api/productivity |
GET / POST |
Productivity metrics per employee and team |
/api/forecast |
GET |
Labor cost projections with seasonal and exogenous variables |
/api/schedules |
GET / POST |
Optimized schedule with labor compliance limits |
/api/burnout |
GET |
Burnout risk scores and auto-trigger gig-worker alerts |
/api/chat |
POST |
Conversational NLP Copilot via CrewAI |
/api/mobile/clock-in |
POST |
Mobile endpoint for employee clock-in with geofencing |
/api/system |
GET |
System status, AI mode (llm or offline), version |
| Component | Technology |
|---|---|
| Backend | FastAPI 0.115 (Python 3.11+) |
| Agent Orchestration | CrewAI 0.80.0 · LangChain 0.3.7 |
| Labor Forecasting | Prophet 1.1.5 |
| Schedule Optimization | PuLP 2.9.0 (linear programming) |
| Burnout Scoring | scikit-learn 1.5.2 |
| Benchmark Embedding | sentence-transformers 3.2.1 |
| AI Narrative | Anthropic Claude (optional) · offline heuristics fallback |
| Database | SQLite (zero-server, local-first) · SQLAlchemy |
| Dashboard | HTML + CSS + JavaScript · Chart.js |
| Testing | pytest · pytest-asyncio · httpx |
Scheduling as optimization, not heuristics. PRISM uses PuLP integer linear programming to find schedules that minimize labor cost while satisfying coverage constraints — not greedy assignment or random search.
Seasonal labor forecasting. Prophet decomposes labor cost time-series into trend and seasonality components, producing forecast intervals that account for known business cycles (holidays, fiscal quarters, peak seasons).
Burnout risk as a lagging-indicator model. The scikit-learn classifier uses a combination of workload intensity, overtime frequency, tenure, and historical absenteeism to score burnout risk before it manifests — not after.
Zero-server dependency. SQLite requires no database server. The full workforce intelligence stack starts with python run.py.
NIST AI RMF 1.0 alignment. Every agent decision, forecast run, and AI-generated narrative is logged with structured metadata following NIST governance principles: validity, reliability, explainability, and human oversight.
PRISM is built for SME operations managers, HR directors, and labor cost analysts who need workforce intelligence without enterprise HRMS costs.
You do not need an operations research background. Register your team, log productivity metrics, and run python run.py. PRISM handles forecasting, schedule optimization, and risk scoring automatically.
git clone https://github.com/afild/PRISM.git PRISM
cd PRISM
pip install -r requirements.txt
python run.pyOpen http://localhost:8004 in your browser.
LLM Mode:
export ANTHROPIC_API_KEY=your_key_here # Linux/macOS
set ANTHROPIC_API_KEY=your_key_here # Windows
python run.pyOffline Mode (default): All analytics — forecasting, schedule optimization, burnout scoring, and KPI computation — operate identically. The CrewAI workforce narrative agent uses structured rule-based report templates.
- Python 3.11 or higher
- Git
git clone https://github.com/afild/PRISM.git PRISM
cd PRISM
python -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env
python run.pypytest tests/ -v| NIST Function | PRISM Implementation |
|---|---|
| GOVERN | MIT License · open audit trail · traceable agent decisions |
| MAP | Workforce analytics domain scoped to SME labor management · documented model assumptions |
| MEASURE | pytest suite · Prophet forecast residuals per run · PuLP solver convergence validation |
| MANAGE | Offline fallback · optimization constraints enforced · burnout alerts require human review |
The AI narrative agent receives only computed aggregate metrics (KPIs, forecast values, optimization results) — never individual employee records or sensitive labor data.
PRISM/
├── app/
│ ├── main.py ← FastAPI app · router · static serving
│ ├── config.py ← Pydantic settings
│ ├── agents/
│ │ ├── orchestrator.py ← CrewAI orchestration across 6 agents
│ │ ├── productivity_tracker.py ← KPI computation per employee and team
│ │ ├── labor_forecaster.py ← Prophet labor cost forecasting
│ │ ├── scheduling_optimizer.py ← PuLP schedule optimization
│ │ ├── burnout_risk.py ← scikit-learn burnout risk classifier
│ │ └── benchmark_lookup.py ← Industry benchmark ingestion and lookup
│ ├── api/
│ │ ├── router.py
│ │ ├── employees.py
│ │ ├── productivity.py
│ │ ├── forecasts.py
│ │ ├── schedules.py
│ │ ├── alerts.py
│ │ ├── chat.py ← NLP Conversational Router
│ │ ├── mobile.py ← Geofenced Mobile Gateway
│ │ └── system.py
│ ├── database/
│ │ ├── db_manager.py
│ │ └── schema.sql ← Tables: employees · productivity_logs · schedules · forecasts
│ └── plugins/
│ └── nova_reader.py ← Read-only NOVA financial context connector
├── data/ ← Benchmark datasets by industry
├── docs/
│ └── images/ ← Architecture diagrams
├── frontend/
│ ├── index.html ← Glassmorphic dashboard
│ ├── styles.css
│ └── app.js ← Chart.js · schedule grid · KPI cards
├── tests/
├── .env.example
├── requirements.txt
└── run.py
Areas where contributions are most needed:
- ATS (Applicant Tracking System) integration for hiring cost modeling
- Multi-location workforce management across time zones
- Real-time productivity API connectors (time-tracking tools)
- Docker Compose setup for zero-dependency deployment
- Compliance Engine: Mathematical constraints in PuLP solver to prevent illegal schedules (e.g. "clopening" penalties).
- Exogenous Forecasting: Added
add_regressorin Prophet for weather/POS data integration. - Conversational Copilot: New
/api/chatendpoint to parse natural language intentions to the CrewAI orchestrator. - Mobile Geofencing: New
/api/mobile/clock-inendpoint preventing remote punch-in fraud. - Gig-Worker Webhooks: Auto-trigger for contingent workers when burnout risk exceeds critical thresholds.
- Full workforce intelligence pipeline: productivity tracking, labor forecasting, schedule optimization, burnout risk, benchmark comparison
MIT License — free to use, adapt, and redistribute.