README.md

---
language: en
license: mit
tags:
  - project-management
  - risk-analysis
  - transformer
  - pytorch
pipeline_tag: text-generation
---

# PMReasoner

PMReasoner is a 125.3M parameter transformer trained from scratch to perform project risk
analysis. It takes a structured project plan (from PMPlanner) and outputs a JSON risk
assessment: overall health (green/yellow/red), top risks, critical path analysis, and
actionable recommendations.

## Model Details

| Property | Value |
|---|---|
| Parameters | 125.3M |
| Architecture | Custom encoder-decoder transformer |
| Training data | 28,000+ PM scenarios with risk annotations |
| Output format | Structured JSON (health, risks, critical_path, recommendations) |
| License | MIT |

## Usage

PMReasoner is part of the PMCore pipeline. Use via the API:

```bash
curl -X POST http://localhost:8765/plan/quick \
  -H "Content-Type: application/json" \
  -d '{"request": "Migrate enterprise from on-prem to cloud, 18 months, $15M budget."}'
```

Response includes both planner and reasoner output:
```json
{
  "planner": {"methodology": "hybrid", "num_tasks": 47, ...},
  "reasoner": {"overall_health": "yellow", "top_risks": ["vendor lock-in", ...], ...}
}
```

## Full Pipeline

See [PMCore on GitHub](https://github.com/snavazio/pmcore) for full documentation.