AutoModeler

3-statement financial models from public data, in seconds.

---

What this does

Type a ticker. Get a fully-linked, projection-ready Excel file — income statement, balance sheet, cash flow, all wired together with native Excel formulas (=B5*B6, not pasted values).

Pulls historical financials from Financial Modeling Prep (FMP), drives projections with a small set of operating assumptions you can tweak in the browser, and downloads as .xlsx.

Why it's interesting

Most "auto-modeling" tools dump numbers into a sheet and call it done. Two things this one does differently:

• Balance sheet plug as a circuit breaker. When the model doesn't balance, most tools silently fudge a row. AutoModeler routes the difference into a cash or revolver plug — but if that plug exceeds a sanity threshold, it throws an error instead of pretending nothing's wrong. You want loud failures here.
• Real Excel formulas, not values. Projected periods are emitted as live xlsxwriter formulas, so opening the file in Excel and editing an assumption actually recalculates the model. Auditors and analysts can trace every cell back to its inputs.

How it works

        FMP API (validated by Pydantic)
              │
              ▼
        Historicals (5y income, balance, cash flow)
              │
              ▼
   Operating drivers ── revenue, margin, capex, WC days
              │
              ▼
        Projection engine (cash / revolver plug)
              │
              ▼
        xlsxwriter — linked .xlsx download

The web UI is a single FastAPI page — no DB, no auth, stateless. Change a driver, see the metrics update, download the file.

Quick start

You'll need an FMP API key (free tier works).

macOS / Linux:

export FMP_API_KEY="your_api_key_here"
pip install -e .
automodeler

Windows (PowerShell):

$env:FMP_API_KEY="your_api_key_here"
pip install -e .
automodeler

Then open http://localhost:8000, enter a ticker (AAPL, MSFT, NVDA), and adjust the base revenue and margin assumptions. The page returns key metrics and a download link for the Excel file.

Development setup

pip install -e ".[dev]"

python-dotenv is included in the dev extras so local .env files are automatically loaded when running the app. In production, set FMP_API_KEY directly as an environment variable.

Running tests

pytest tests/ -v

Balance sheet tolerance

The projection engine validates that the balance sheet balances each period. The allowed plug magnitude defaults to 10% of total assets (threshold_pct=0.10). You can relax this threshold when constructing a FinancialModel:

engine = FinancialModel(history, drivers)
projected = engine.project(horizon=5, threshold_pct=0.15)

If the balance sheet plug exceeds the threshold, a ValueError is raised — this is intentional. It means the model's assumptions are internally inconsistent and the output should not be trusted.

Repo structure

src/automodeler/
├── app.py           # FastAPI routes + Jinja templates
├── fetch.py         # FMP API client, Pydantic-validated
├── model.py         # 3-statement model assembly + plug logic
├── export.py        # xlsxwriter formula emission
├── templates/       # browser UI
└── static/
tests/
├── test_fetch.py
└── test_model.py

Known limitations

• Tax & interest use static assumptions. A real debt schedule is the next obvious step.
• Share buybacks are ignored — equity is held flat outside retained earnings.
• Working capital drivers (DSO, DIO, DPO) are standard, but software tickers really want deferred revenue modeled separately.
• No DCF or LBO — this is a 3-statement model only; valuation overlays are out of scope for v1.

When you'd use this

• Vetting an investment thesis quickly without rebuilding a model from scratch
• Teaching the mechanics of 3-statement linkage to analysts or students
• Spot-checking a sell-side model against a clean baseline

License

MIT — see LICENSE.