FinTag - AI-Powered Financial Analysis Tools

Two AI-powered financial analysis tools built on SEC EDGAR data.

Tools

FinTag - XBRL Tag Classifier

Predicts the most likely US-GAAP XBRL tag for any financial line item.

• Training data: 1.75M label-tag pairs extracted from 68 SEC quarterly datasets (44M raw rows)
• Model: QLoRA fine-tuned Qwen3-4B (r=16, alpha=32, 3 epochs on A100-40GB)
• Baseline: Bigram similarity matching against 1,000-tag vocabulary (86.4% coverage)
• Input: Financial line item text + statement type (IS/BS/CF/EQ/CI) + optional SIC code

FinAnomaly - Financial Ratio Anomaly Detector

Flags companies whose financial ratios deviate significantly from industry peers.

• Data: 74,993 company-period records across 7,553 companies, 71 industries (2022-2024)
• Ratios: 10 financial ratios (margins, ROA/ROE, leverage, liquidity, R&D intensity, AR turnover)
• Method: IQR-based scoring by 2-digit SIC code (robust against fat-tailed financial data)
• Source: SEC EDGAR Financial Statement Data Sets (num.txt + sub.txt)

Quick Start

pnpm install
pnpm dev

Open http://localhost:3000 (or the port shown in terminal).

Data Pipeline

The scripts/ directory contains the Python data pipeline:

68 SEC quarterly zips -> extract_fintag_data.py -> 1.75M SFT training pairs
                      -> build_finanomaly_db.py -> DuckDB (74K company-period ratios)
                      -> modal_fintag_train.py  -> QLoRA adapter (Modal A100-40GB)

To rebuild from scratch:

./scripts/run_all.sh         # extract + build DB
./scripts/run_all.sh --train # also run Modal training

Architecture

• Next.js 16 App Router with Tailwind CSS
• API routes read pre-exported JSON data server-side (no database dependency)
• FinTag baseline uses bigram Dice coefficient for fuzzy matching
• FinAnomaly uses IQR-based z-scores (more robust than standard deviation for financial data)

Tech Stack

Python (data pipeline) + TypeScript/Next.js (demo UI) + Modal (GPU training)