Kestrel Prediction Engine
AI-powered project simulation and forecasting.
Kestrel is Osprai's prediction engine. It runs a multi-agent simulation system that models construction and AV project dynamics to forecast timelines, costs, and risks.
Beta feature
Kestrel is currently in beta. Access it from Simulations in the sidebar.
What Kestrel does
Kestrel runs Monte Carlo simulations of your project using AI agents that model real-world behaviour: contractors, engineers, suppliers, and project managers each making decisions based on their goals and constraints.
Key capabilities:
- Timeline forecasting. Probability distributions for project completion, not just single-point estimates
- Cost modelling. Budget scenarios accounting for delays, change orders, and cascading impacts
- Risk identification. Surfaces risks before they materialise based on project characteristics
- Delay cascade analysis. Models how a delay in one trade propagates to others
- What-if scenarios. Change variables and see how outcomes shift
How it works
Seed your simulation
Describe your project (type, location, budget, timeline, procurement model) or upload project documents for Kestrel to extract the parameters automatically.
Configure parameters
Set the number of simulation runs, project phases, and key variables. Kestrel provides sensible defaults based on project archetype (office tower, venue fit-out, residential, etc.).
Run the simulation
Kestrel runs hundreds of simulated project timelines, each with AI agents making realistic decisions. The results converge into probability distributions for duration and cost.
Review results
Explore the simulation report:
- Duration forecast. P50, P80, P90 completion dates
- Cost forecast. Expected budget range with confidence intervals
- Risk register. Identified risks ranked by impact probability
- Critical path. Which phases and trades drive the timeline
- Agent behaviour. What decisions led to delays or cost overruns
Project archetypes
Kestrel includes built-in archetypes for common project types:
| Archetype | Typical duration | Phases |
|---|---|---|
| Office tower | 18-36 months | Foundation, Structure, Envelope, MEP, Fit-out |
| Venue / Arena | 12-24 months | Shell, MEP, AV rough-in, AV trim, Commissioning |
| Residential complex | 14-30 months | Site, Structure, Envelope, Interior, Landscape |
| Renovation | 6-18 months | Demo, Structure, MEP, Finishes |
| AV fit-out | 3-8 months | Design, Procurement, Rough-in, Trim, Commissioning |
Each archetype comes with calibrated parameters based on industry data.
Simulation vs. reality
Forecasts, not guarantees
Kestrel provides probabilistic forecasts based on simulation models. Results should inform decision-making alongside professional experience and site-specific knowledge. The model improves as you provide more project-specific data.
Tips
- More runs = more accuracy. 100+ runs give stable probability distributions; 3-10 runs are useful for quick directional estimates
- Ground in your documents. Upload your project schedule, budget, and specs to give Kestrel real data to work with
- Compare scenarios. Run multiple simulations with different assumptions (e.g., with and without a second shift) to quantify the impact
- Check the agent log. Understanding why simulated agents made certain decisions helps you anticipate real-world behaviour