Audit Your Financial Models with AI — Like Grammarly for Excel.
ModelGuard catches formula errors, hidden risks, and suspicious patterns in financial models. Built for FP&A, investment banking, audit, and accounting professionals who need confidence in their spreadsheet outputs.
Financial models drive billion-dollar decisions. A single formula error, hidden assumption, or inconsistent cell can cascade into flawed valuations, misallocated capital, or regulatory exposure. Most model audits are manual, time-consuming, and error-prone.
ModelGuard automates the audit — scanning every cell, every sheet, every hidden element — and produces a structured, explainable report in seconds.
- Detects
#REF!,#DIV/0!,#VALUE!errors across all sheets - Flags hardcoded constants inside formula cells
- Identifies inconsistent neighboring formula patterns
- Reveals hidden sheets, hidden rows, hidden columns
- Detects external workbook links that can break
- Identifies circular reference patterns
- Flags suspicious revenue growth (>100% period-over-period)
- Detects margin jumps (>20% period change)
- Warns on negative cash flow and empty assumption sections
Every issue includes:
- Plain-English description of what's wrong
- Why it matters for financial modeling
- Suggested fix with context
- Professional PDF audit reports
- CSV export for issue tracking
- Clean, recruiter-friendly dashboard
| Layer | Technology |
|---|---|
| Frontend | Next.js 14, TypeScript, Tailwind CSS, shadcn/ui, Framer Motion, Recharts |
| Backend | FastAPI, Python 3.11+, openpyxl, pandas |
| Database | SQLite (MVP), audit history tracking |
| AI | Mock AI layer (pluggable to OpenAI/OpenRouter) |
| Design | Dark fintech theme (inspired by Bloomberg Terminal, Linear, Mercury) |
┌─────────────────────────────────────────┐
│ Next.js Frontend │
│ Landing → Upload → Audit Dashboard │
│ (shadcn/ui + Framer Motion) │
└──────────────┬──────────────────────────┘
│ /api/upload, /api/demo
┌──────────────▼──────────────────────────┐
│ FastAPI Backend │
│ Parser (openpyxl) → Audit Engine → DB │
│ 12 detection rules │
└─────────────────────────────────────────┘
# Clone
git clone <repo-url> && cd modelguard-ai
# Backend
cd backend
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --port 8000
# Frontend (new terminal)
cd frontend
npm install
npm run devOpen http://localhost:3000 — click Try Demo for an instant audit preview.
- Landing Page — Dark hero with CTA and feature cards
- Upload — Drag-and-drop your .xlsx file
- Processing — Animated audit steps (Reading → Scanning → Detecting → Generating)
- Dashboard — Health score, severity breakdown, issues table, AI summary
- Export — PDF report or CSV issue list
{
"model_score": 48,
"summary": "Detected 7 issues. Model is high risk...",
"issues": [
{
"severity": "critical",
"sheet": "Forecast",
"cell": "H42",
"category": "Formula Error",
"description": "#REF! in revenue forecast",
"why_it_matters": "Broken formulas cascade into valuation outputs",
"suggested_fix": "Trace precedent cells and restore reference"
}
],
"severity_breakdown": { "critical": 1, "high": 3, "medium": 3, "low": 0 }
}-
Built ModelGuard AI, a full-stack financial model auditing platform that detects 12 categories of Excel errors including formula integrity, hidden risks, and business logic anomalies using Python (FastAPI, openpyxl, pandas) and TypeScript (Next.js, Tailwind CSS).
-
Designed a dark fintech dashboard inspired by Bloomberg Terminal and Linear, featuring animated health scores, severity breakdowns, interactive issue tables with detail modals, and professional PDF/CSV export — deployed as a portfolio-ready demo with mock data.
-
Engineered a 12-rule audit engine that scans Excel workbooks for
#REF!/#DIV/0!errors, hidden sheets/rows/columns, circular references, external links, suspicious revenue growth (>100%), margin jumps, and negative cash flow patterns — producing structured JSON output with plain-English explanations. -
Integrated AI-style explanations where each detected issue includes context-aware descriptions of why it matters for financial modeling professionals and actionable suggested fixes — designed to plug into OpenAI/OpenRouter for production use.
- OpenAI/OpenRouter integration for real AI explanations
- User accounts and audit history
- Multi-file model comparison
- Excel formula dependency graphing
- Custom audit rule builder
- Slack/Teams notification integration
- Codex CLI — Implementation and code review
- OpenArt — Visual assets (hero, logo, illustrations)
- Hermes Agent — Project orchestration and task delegation
cd frontend
vercel login # Interactive browser OAuth required
vercel --prod # Deploys with vercel.json configSet environment variable: NEXT_PUBLIC_API_URL=https://your-backend.onrender.com
cd backend
# Render: connect GitHub repo → set build command: pip install -r requirements.txt
# Start command: uvicorn main:app --host 0.0.0.0 --port $PORTCORS configured in main.py — update allow_origins with production frontend URL.
The in-memory limiter trusts request.client.host by default and ignores raw
X-Forwarded-For values from untrusted peers. If you operate behind a proxy
whose source network you control, set MODEL_GUARD_TRUSTED_PROXY_CIDRS to a
comma-separated CIDR allowlist, for example:
MODEL_GUARD_TRUSTED_PROXY_CIDRS=10.0.0.0/8,2001:db8:1234::/48Do not use a universal trust value such as *. Uvicorn proxy-header trust
should be configured consistently with --forwarded-allow-ips or the
FORWARDED_ALLOW_IPS environment variable. The local limiter is bounded and
single-instance; use a shared external limiter before horizontally scaling.
# Backend
cd backend && source venv/bin/activate && uvicorn main:app --port 8000
# Frontend (new terminal)
cd frontend && npm run dev| Check | Result |
|---|---|
| 9 API endpoints | All 200 OK |
| Frontend build | 11/11 pages, zero errors |
| accessibility | rel="noopener noreferrer" on external links |
| Security audit | No API keys exposed, GAAP-only data, source URLs verified |
- Screen capture MCP tool broken — screenshots require manual browser capture
- Backend deploy requires Render/Railway account (browser setup)
- Vercel deploy requires
vercel login(browser OAuth) - Real peer data only for SaaS industry (10 companies)
- No authentication — demo-only deployment
- sound MCP broken — no audio feedback on flex
- Uploaded audit results are stored client-side in sessionStorage for the demo.
- Refreshing /results clears the displayed report.
- Public demo should not be used with confidential client files.
- Production deployment would require authenticated storage, retention controls, and stronger security review.
- The backend does not permanently store uploaded workbook contents after the audit completes.
Built as a fintech portfolio project demonstrating full-stack engineering, financial domain knowledge, and UI/UX design.
- Frontend: https://modelguard-ai.vercel.app
- Backend API: https://modelguard-ai.onrender.com
- Demo Dashboard: https://modelguard-ai.vercel.app/demo
- Benchmark: https://modelguard-ai.vercel.app/benchmark