Where macro signal meets machine reasoning.
ProjectCortex is an institutional-grade forecasting stack — Snowflake-native pipelines, economic signal extraction, anomaly detection, and an LLM reasoning layer — designed from the ground up for S&P, BankFocus and FDIC data.
- 7
- Pipeline modules
- 2
- First-party datasets
- LLM
- Reasoning layer
- FastAPI
- Service surface
The engine
Seven modules, one strategic surface.
Every layer is independently runnable and independently testable — wired together into a single forecasting brain that thinks in macro signals and bank fundamentals.
Snowflake integration
First-class warehouse adapters, secure key-pair auth, and incremental ingestion built for BankFocus and FDIC scale.
Forecasting pipeline
Composable models — classical, gradient-boosted, and deep — orchestrated end-to-end with reproducible backtests.
Economic signal extraction
Macro regime features, yield-curve dynamics, credit spreads and bank-call-report indicators distilled to model-ready signals.
Anomaly detection
Statistical and learned detectors flag balance-sheet drift, deposit shocks and reporting irregularities in real time.
LLM reasoning layer
A grounded reasoning surface that narrates forecasts, weighs counterfactuals and cites the underlying signal evidence.
FastAPI service
Production-grade async API with typed contracts, OpenAPI docs, and a Dockerized footprint ready for any cluster.
Use cases
Built for the institutions that move markets.
Cortex is deployed where forecasts carry real capital weight — regulators, strategy desks, banks and quant teams that need reasoning, not just numbers.
Systemic risk surveillance
Monitor deposit flight, capital adequacy drift and concentration risk across thousands of institutions — with anomaly alerts cited back to the underlying call-report line items.
Macro-aware forecasting
Forecast loan growth, NIM compression and credit quality under explicit macro regimes. The LLM layer narrates each forecast and weighs counterfactual rate paths.
Peer benchmarking & planning
Benchmark balance-sheet performance against custom peer cohorts, project quarterly trajectories, and stress-test scenarios end-to-end on your own warehouse.
Alpha from regulatory data
Turn FDIC and BankFocus filings into model-ready signals — regime features, anomaly scores and reasoned theses ready for portfolio construction and credit underwriting.
Architecture
From raw call report to reasoned forecast.
A linear, auditable flow. Every stage emits structured artifacts the next stage consumes — and the LLM cites them by name.
S&P, BankFocus · FDIC
Raw call reportsSnowflake
Warehouse + stagingFeature pipeline
Signals · regimesForecast models
Ensembles · backtestsAnomaly layer
Drift · shocksLLM reasoning
Narrate · citeFastAPI
Typed contractsDatasets
Built for BankFocus, FDIC and S&P from day one.
Cortex isn't a generic forecasting toolkit retrofitted to finance. The schemas, feature extractors and reasoning prompts are co-designed with three of the deepest banking and credit datasets in the world.
BankFocus
Global bank financials, ownership and credit indicators — normalized into a forecasting-ready feature store.
FDIC Call Reports
Quarterly call-report extracts streamed into Snowflake with versioned schemas and reporting-period diffs.
S&P Capital IQ / Market Intelligence
Credit ratings, issuer fundamentals, and market-implied default probabilities — integrated for peer benchmarking and risk overlay.
Roadmap
A path to institutional-grade strategic forecasting.
Foundation
- Snowflake adapter
- BankFocus + FDIC ingestion
- FastAPI scaffold
- Dockerized stack
Signal & forecast
- Macro regime features
- Ensemble forecast models
- Backtesting harness
- Anomaly detectors
Reasoning
- LLM reasoning layer
- Cited counterfactuals
- Scenario simulator
- Analyst-grade reports
Institutional
- SOC2-ready controls
- Multi-tenant isolation
- Streaming ingestion
- Strategy-desk integrations
Deploy Cortex inside your institution.
ProjectCortex is in private preview with a select group of strategy desks and quant teams. Request access to run the stack on your own warehouse.