FinSight Planner
Financial Planning & Scenario Intelligence
FinSight Planner extends the FinSight principle into planning: calculate with deterministic financial models, explain with AI. It supports variance analysis, assumption-driven forecasts and what-if scenarios for decision support.
Phynapse research & engineering project · Synthetic data
The business problem
Planning decisions depend on assumptions: growth rates, cost changes, timing. Exploring how those assumptions change outcomes is often slow, spreadsheet-bound and hard to explain to stakeholders.
Introducing AI into planning raises the same risk as in reporting: figures that look plausible but are not computed from the model.
What we engineered
Every figure comes from deterministic Python/Pandas financial models. An optional language model only turns questions into a validated analysis plan and rewrites computed evidence into prose; any figure it produces that is not present in the evidence is rejected.
What the system does
Variance analysis
Actual versus budget and forecast, structured for review.
Assumption-based modelling
Forecasts driven by explicit, inspectable assumptions.
What-if scenarios
Alternative scenarios branched from a baseline for comparison.
Evidence-checked explanations
Narrative explanations validated against computed evidence.
Transparent explanation source
Responses state whether prose came from the model or the deterministic fallback.
Deterministic mode
Runs without any model credentials.
Design decisions
- 01
Calculate with deterministic financial models; explain with AI.
- 02
Model output is treated as untrusted: numbers not found in the evidence are rejected.
- 03
API keys stay server-side and are never returned by the API.
- 04
Scenario methodology, financial model and AI architecture are documented in the repository.
How information moves through the system
- 1
Baseline
Actuals, budget and forecast data loaded read-only.
- 2
Assumptions
Explicit drivers define each scenario.
- 3
Deterministic model
Python / Pandas computes projected outcomes.
- 4
Scenario comparison
Alternatives evaluated against the baseline.
- 5
Explanation
Optional LLM prose, validated against evidence.
Built with
- Python 3.11+
- FastAPI
- Pandas
- React + Vite
- TypeScript
- pytest
- OpenAI-compatible LLM (optional)
As documented in the project repository.
What this project is, and is not
- Engineered decision-support demonstration using a fictional company and synthetic data.
- Not an accounting system, ERP or audited reporting source.
- Sparse sample data; budget and forecast coverage is partial.
- Scenarios and evidence are held in memory, with no authentication or persistence.
Review the engineering in the open.
The full source, documentation and tests for FinSight Planner are available on GitHub.