Veritas: Autonomous AI Due Diligence Terminal

Veritas is a local-first, multi-agent AI system designed to perform adversarial due diligence on startup pitch decks. Unlike standard chatbots, Veritas operates as an Investigative Operating System—treating startup claims as hypotheses to be verified against external reality.

🚀 Key Features

• Zero-Trust Claim Extraction: Automatically identifies "Hard Claims" (revenue, growth, partners) while filtering out marketing fluff.
• Adversarial Research: Dispatches agents to find evidence that contradicts startup claims using Tavily & Serper.
• The Skeptic Agent: Models "expected visibility" to detect "the dog that didn't bark"—identifying when evidence is suspiciously absent.
• Bayesian Credibility Engine: Mathematically propagates trust and decay throughout the company narrative.
• Investigative Terminal: A multi-panel Dashboard built with Next.js for real-time trace inspection and source-grounded reasoning.

---

🛠 Tech Stack

• Reasoning: Ollama (qwen2.5:8b)
• Orchestration: LangGraph (Stateful, cyclical multi-agent workflows)
• Backend: FastAPI (Async IO & WebSockets)
• Frontend: Next.js 14, TailwindCSS, shadcn/ui, Zustand
• Retrieval: FAISS (Local session-based vector storage)
• Search: Tavily AI Search

---

🏁 Getting Started

1. Prerequisites

• Python 3.12+ (via pyenv)
• Node.js 18+
• Ollama installed and running

2. Setup Models

Pull the required local models for reasoning and embeddings:

ollama pull qwen3:8b
ollama pull nomic-embed-text

3. Environment Configuration

Create a .env file in the root directory:

TAVILY_API_KEY=your_api_key_here

4. Installation

# Activate your environment
pyenv activate veritas

# Install Python dependencies
pip install -r requirements.txt

# Install Frontend dependencies
cd frontend
npm install

---

🏃 Running the System

Step 1: Start the Backend

# In the root directory
python -m app.main

The API will be available at http://localhost:8000

Step 2: Start the Frontend

# In a new terminal
cd frontend
npm run dev

The Terminal UI will be available at http://localhost:3000

---

📂 Project Structure

├── app/
│   ├── agents/          # Extractor, Researcher, Analyst, Skeptic, Synthesis
│   ├── api/             # FastAPI routes & WebSocket manager
│   ├── core/            # Belief Engine & Bayesian logic
│   ├── orchestration/   # LangGraph workflow definitions
│   ├── retrieval/       # FAISS & local vector store service
│   ├── schemas/         # Pydantic V2 Epistemic models
│   └── services/        # PDF Parsing & Ingestion logic
├── docs/                # Technical & UI/UX Specifications
├── frontend/            # Next.js 14 Investigative Terminal
└── scripts/             # Standalone test scripts for agents

---

⚖️ Epistemic Philosophy

Veritas is built on Institutional Skepticism.

• Supported: Evidence corroborates the claim.
• Contradicted: Reality conflicts with the narrative.
• Suspiciously Absent: A high-impact claim has zero external footprint.
• Ambiguous: Conflicting or low-quality data.

---

📜 Documentation

For deep-dives into the architecture, see:

• docs/technical-specification.md
• docs/ui-ux-specification.md
• TASKS.md

---

Built for the future of rigorous financial intelligence.