🚀 Quick Start Guide | ⚡ Lightning AI Setup | 🌐 Deployment Guide
Event: InnVedX Code Hackathon, Utkarsh 2026
Venue: BBD University, Lucknow
Date: 26 February 2026
Project Name: PRAMAAN (प्रमाण)
Problem Statement: Democratizing Financial Intelligence: Decoding Corporate PR vs. Financial Reality using LLM (OPEN INNOVATION)
- Aamin Simmi Singh, BBD University
- Harshit Tiwari, BBD University
- Vansh Dixit, BBD University
PRAMAAN stands for Professional Reports & Analysis on Money, Assets And Net-gains.
Every year, retail investors lose billions of dollars because they trust the glossy, optimistic PR spin in a company's Annual Report, completely missing the catastrophic red flags hidden deep within the auditor's notes, cash flow statements, and balance sheets.
PRAMAAN bridges this gap. It acts as an autonomous forensic accountant that reads between the lines of corporate filings to separate the Projected Story from the Actual Reality.
PRAMAAN doesn't just summarize PDFs; it actively cross-references claims. When management boasts about "record-breaking revenue" or "debt-free operations," PRAMAAN scans the statutory filings, auditor qualifications, and accounting policies to verify if those claims are true, or if they are smoke and mirrors.
For every analysis, PRAMAAN delivers a brutal, easy-to-understand breakdown:
- The Projected Story: What the management wants you to believe.
- The Actual Reality: The specific financial red flags (or green flags) hidden in the data.
- The Layman's Conclusion: A completely jargon-free explanation of the company's actual financial health.
Standard AI models fail at forensic accounting because they are trained on generic text and become easily tricked by corporate jargon.
PRAMAAN is powered by a custom-built, highly lethal Fine-Tuning Dataset engineered specifically for this hackathon. Our dataset trains the model to recognize the exact mechanics of accounting manipulation, including:
- The "Adjusted EBITDA" Illusion: Ignoring hundreds of crores in employee stock compensation (ESOPs).
- Ghost Receivables & Factoring: Hiding dead loans to bankrupt subsidiaries or faking cash flow by selling invoices at a discount.
- Capitalizing Expenses: Shifting everyday marketing and operational bills onto the balance sheet to artificially inflate current profits.
- The "Going Concern" Crisis: Matching boastful turnaround projections against an auditor's official warning of imminent bankruptcy.
We didn't just train PRAMAAN on the bad guys. We also included a "Control Group" of elite Green Flag synthetic data, teaching the AI what radical financial transparency and honest accounting actually look like, ensuring it doesn't just blindly flag every company it reads.
PRAMAAN democratizes elite-level financial analysis. Whether it is a telecom giant drowning in debt, a new-age tech startup burning cash, or a massive conglomerate hiding liabilities in shell companies, PRAMAAN gives everyday investors the power of a Wall Street short-seller in seconds.
Truth is in the numbers. PRAMAAN finds it.
- Node.js v18+
- Python 3.12+
- uv -
curl -LsSf https://astral.sh/uv/install.sh | sh
Terminal 1 - Backend:
cd pramaan/backend
uv sync
uv run uvicorn main:app --host 0.0.0.0 --port 8000 --reloadTerminal 2 - Frontend:
cd pramaan/frontend
npm install
npm run devOpen http://localhost:5173 in your browser!
📖 Full instructions: STARTUP.md
pramaan/
├── backend/ # FastAPI backend (Python + uv)
├── frontend/ # React + Vite frontend
├── lightning_ai_server/ # Model deployment for Lightning AI
├── datasets/ # Training datasets
├── STARTUP.md # Complete startup guide
└── LIGHTNING_AI_SETUP.md # Lightning AI deployment guide
| Document | Description |
|---|---|
| STARTUP.md | Complete local development setup |
| LIGHTNING_AI_SETUP.md | Deploy model on Lightning AI with T4 GPU |
| lightning_ai_server/ | Server code for Lightning AI |
| lightning_ai_server/KEEP_RUNNING.md | Keep server running 24/7 |
- Frontend: React 19, Vite, TailwindCSS, TypeScript
- Backend: FastAPI, SQLAlchemy, Firebase Auth
- AI Model: Fine-tuned Llama-3 via Ollama
- Deployment: Lightning AI (T4 GPU)