Financial Report Intelligence System — Case Study

Overview

Rather than functioning as a chatbot, FRIS transforms unstructured financial documents — income statements, balance sheets, cash flow statements — into structured metrics, insights, and business intelligence through a specialised multi-agent architecture: document processing, metrics extraction, financial calculations, data quality, financial analysis, insight generation, risk assessment, visual blueprints, narrative synthesis, and export.

The guiding principle: every insight should be traceable to source evidence. Python performs the calculations; AI provides the interpretation — evidence-linked risk assessment and anomaly detection, trend and variance analysis across reporting periods, and AI-generated management commentary, with model-agnostic support for local models (Ollama, Phi-3 Mini, Llama 3.1) or hosted providers (OpenAI, Anthropic, Google Gemini).

Key Results

10

Specialised agents, one per task in the pipeline

21+

Financial metrics & ratios calculated automatically

37 tests

Passing, validated against Apple, Amazon & Colgate annual reports

~1s–48s

Processing time — text-native reports to OCR-required documents

Tech Stack & Skills Demonstrated

Multi-Agent AI Architecture

Model-agnostic (Ollama/Phi-3/Llama 3.1, OpenAI, Anthropic, Gemini), 10 specialised agents, evidence-linked outputs

Financial Statement Analysis

Multi-period extraction (income, balance sheet, cash flow), ratio calculation, trend & variance analysis

Data Engineering

PyMuPDF / OpenPyXL document parsing, SQLite, PostgreSQL, ChromaDB storage

Reporting

AI-generated executive commentary, Power BI dashboard specification generation

Project README

Pulled live from the GitHub repository — always in sync with the source.

Explore the Code

Full architecture, agent design, and test suite are on GitHub.

GitHub ↗ Visual Write-up on Notion ↗ ← Back to Portfolio