Data Schema

Financial Intelligence Platform

Five interconnected tables in DuckDB covering market prices, banking transactions, loan portfolios, fraud detection, and automated data quality logs. PK = primary key · FK = logical foreign key reference.

DuckDB Python FastAPI :8765 yfinance Local Llama (Ollama) LangChain NL-to-SQL

~26k

Market data rows (real OHLCV, yfinance)

500k

Synthetic banking transactions

200k

Synthetic loan records

~850

Fraud detection events

ETL + DQ Pipeline

Data flows from sources through quality validation into the warehouse, then queried by the NL-SQL agent.

yfinance API

Synthetic Generators

→

DQ Validation

→

DuckDB Warehouse

→

Lineage Tracker

→

NL-SQL Agent (Llama)

→

FastAPI :8765

Warehouse Schema

Five tables. PK = primary key · FK = logical foreign key (not enforced at DB level).

market_data

Real OHLCV · 20 tickers · 5 years

idPKBIGINT

symbolVARCHAR(10)

trade_dateDATE

openDOUBLE

highDOUBLE

lowDOUBLE

closeDOUBLE

adj_closeDOUBLE

volumeBIGINT

ingest_tsTIMESTAMP

transactions

500k synthetic banking txns · 3yr

transaction_idPKVARCHAR(36)

account_idVARCHAR(20)

customer_idVARCHAR(20)

txn_dateTIMESTAMP

amountDOUBLE

merchant_nameVARCHAR(100)

merchant_categoryVARCHAR(50)

channelVARCHAR(20)

currencyVARCHAR(3)

is_flaggedBOOLEAN

data_quality_flagVARCHAR(20)

ingest_tsTIMESTAMP

loans

200k consumer loans · 7yr history

loan_idPKVARCHAR(36)

member_idVARCHAR(20)

issue_dateDATE

loan_amountDOUBLE

funded_amountDOUBLE

interest_rateDOUBLE

term_monthsINTEGER

gradeVARCHAR(2)

sub_gradeVARCHAR(2)

purposeVARCHAR(50)

annual_incomeDOUBLE

dtiDOUBLE

fico_lowINTEGER

fico_highINTEGER

home_ownershipVARCHAR(20)

loan_statusVARCHAR(30)

delinq_2yrsINTEGER

ingest_tsTIMESTAMP

fraud_events

Detection events on flagged transactions

event_idPKVARCHAR(36)

transaction_idFKVARCHAR(36)

detected_atTIMESTAMP

fraud_typeVARCHAR(40)

confidence_scoreDOUBLE

model_versionVARCHAR(20)

is_confirmedBOOLEAN

resolved_atTIMESTAMP

resolution_noteVARCHAR(255)

ingest_tsTIMESTAMP

data_quality_log

Automated DQ check results written after every ingestion run

run_idPKVARCHAR(36)

dataset_nameVARCHAR(50)

run_tsTIMESTAMP

total_checksINTEGER

passed_checksINTEGER

failed_checksINTEGER

success_rateDOUBLE

ge_result_pathVARCHAR(255)

ingest_tsTIMESTAMP

Table Relationships

fraud_events references transactions logically. data_quality_log validates all 4 data tables. market_data and loans are independent.

API Endpoints (port 8765)

No authentication required. NL-to-SQL powered by local Llama via Ollama.

POST /query Ask a plain-English question. Returns generated SQL + rows.

GET /query/schema Browse warehouse tables and row counts.

GET /query/dq Latest automated data quality check results per dataset.

GET /query/lineage/{'{'}table{'}'} Data lineage trace from source to warehouse for a given table.

GET /health Health check. Returns {"{"}"status":"ok","port":8765{"}"}.