Financial Report AI

Financial Report AI turns an uploaded company document into a structured, equity-research-style PDF report. It combines document parsing, LLM-based extraction, chart generation, and a Playwright-rendered HTML/CSS report template in one local web workflow.

Demo workflow

1. Open the Next.js app at http://localhost:3000.
2. Enter a company name and upload a PDF, TXT, or CSV document.
3. The FastAPI service extracts source text and tables, then asks the configured LLM for structured report data.
4. The service generates charts, renders the teal research-report PDF, and returns a preview and download link.

The report gracefully shows Not Available where the source document does not contain a requested metric.

Architecture

Document upload
    -> parser (PDF / TXT / CSV)
    -> LLM JSON extraction
    -> normalized report data
    -> matplotlib charts + Jinja report template
    -> Playwright / Chromium PDF
    -> browser preview and download

See architecture.md for component boundaries and data persistence details.

Tech stack

• Frontend: Next.js 15, React, TypeScript, Tailwind CSS, React Hook Form, Axios
• API: FastAPI, Pydantic, Uvicorn
• Parsing: PyMuPDF, pdfplumber, pandas
• AI extraction: OpenAI-compatible Chat Completions API or Mistral
• Report generation: Jinja2, matplotlib, Playwright/Chromium

Prerequisites

• Python 3.11+
• Node.js 18+
• An OpenAI or Mistral API key

Run locally

Open two terminals from the repository root.

1. Configure and start the backend

cd backend
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python -m playwright install chromium
cp .env.example .env

Edit backend/.env, choose exactly one provider, and add its API key. For OpenAI:

PROVIDER=openai
OPENAI_API_KEY=your_api_key
MODEL=gpt-4.1

For Mistral:

PROVIDER=mistral
MISTRAL_API_KEY=your_api_key
MISTRAL_MODEL=mistral-large-latest
MISTRAL_API_BASE=https://api.mistral.ai

Start the API:

uvicorn main:app --reload --host 0.0.0.0 --port 8000

The API is served at http://localhost:8000/api.

2. Start the frontend

cd frontend
npm install
npm run dev

Open http://localhost:3000.

API

Endpoint	Purpose
POST /api/upload	Accepts a multipart file, parses it, extracts structured report data, and returns a job_id.
POST /api/generate	Accepts { "job_id": "..." }, generates chart assets and the PDF, and returns a download path.
GET /api/download/{job_id}	Streams the generated PDF.

Example API flow:

curl -F "file=@company-results.pdf" http://localhost:8000/api/upload
curl -X POST http://localhost:8000/api/generate \
  -H "Content-Type: application/json" \
  -d '{"job_id":"<job-id-from-upload>"}'
curl -OJ http://localhost:8000/api/download/<job-id>

Project structure

backend/
  app/api/routes.py                 # upload, generation, and download endpoints
  app/services/parser.py            # PDF, TXT, and CSV parsing
  app/services/extractor.py         # LLM prompting and response normalization
  app/services/chart_generator.py   # matplotlib chart assets
  app/services/pdf_generator.py     # Jinja + Playwright PDF generation
  app/templates/report.html/.css    # research-report layout
  .env.example                      # provider configuration template
frontend/
  app/page.tsx                      # upload, progress, preview, and download UI

Validation

Before sharing changes, run:

# backend, with the virtual environment activated
python -m compileall -q app

# frontend
cd ../frontend
npm run build

Scope and limitations

• The application is a report-generation MVP, not investment advice or a valuation engine.
• The LLM is instructed to extract only source-supported information, but generated output should be reviewed before any professional use.
• Jobs and generated assets are stored locally for the current implementation; there is no authentication, database, or object storage layer.
• The frontend is configured for the local API at http://localhost:8000; update the client and CORS policy before deployment.

Repository hygiene

Do not commit backend/.env, virtual environments, generated reports, job stores, or logs. These are excluded by .gitignore.