📊 insideX: AI-Powered Stock Anomaly Detector

insideX is an AI-driven platform that monitors financial news, detects anomalies, and provides threat-level insights (0–10) for stocks in real-time. It combines Perplexity AI for latest news, Pathway for memory & vector retrieval, Claude AI for anomaly detection, and MongoDB for persistence.

---

🚀 Features

• 📡 Live Stock News Fetching via Perplexity API
• 🧠 Real-time Vector Memory using Pathway (RAG-style retrieval)
• 💾 Persistent Storage of raw news in MongoDB
• 🔍 Anomaly Detection powered by Claude 3.5 Sonnet
• ⚖️ Threat Level Scoring (0–10) with reasoning based on past cases
• 💬 Follow-up Questions (Chat-style queries for deeper insights)
• ⚡ FastAPI + Uvicorn backend with REST API endpoints

---

🏗️ System Architecture

        ┌──────────────┐
        │   User Query │
        └──────┬───────┘
               │
         /analyze (FastAPI)
               │
   ┌───────────┴─────────────┐
   │                         │
Perplexity API         Pathway Memory (Vector DB)
 (latest news)         + MongoDB (persistent store)
   │                         │
   └───────────┬─────────────┘
               │
         Claude 3.5 Sonnet
     (Anomaly detection + summary)
               │
      Threat Score (0–10) + Report

---

🛠️ Tech Stack

• Backend Framework: FastAPI (Python)
• Vector Database / Memory: Pathway (real-time indexing + retrieval)
• Embeddings Model: BGE-base-en-v1.5 (Hugging Face)
• LLM Reasoning: Claude 3.5 Sonnet (Anthropic)
• External Data Source: Perplexity API
• Database: MongoDB (stores raw news docs)
• Server: Uvicorn (ASGI for FastAPI)
• Config & Secrets: dotenv (.env)
• Frontend (planned): React.js + TailwindCSS

---

📂 Project Structure

insideX/
│── backend/
│   ├── main.py          # FastAPI app entry
│   ├── llm.py           # LLM orchestration (Perplexity + Claude + Threat model)
│   ├── vector_store.py  # Pathway memory & retrieval
│   ├── embeddings.py    # Embedding model (BGE)
│   ├── db.py            # MongoDB helper (save/retrieve news)
│   └── requirements.txt # Python dependencies
│
│── frontend/ (planned)  # React.js + Tailwind UI
│
│── .env.example         # Sample env file
│── README.md            # Documentation

---

⚙️ Setup Instructions

1️⃣ Clone Repository

git clone https://github.com/your-username/insideX.git
cd insideX

2️⃣ Setup Python Environment

python3 -m venv .venv
source .venv/bin/activate   # Linux/Mac
# OR
.venv\Scripts\activate      # Windows PowerShell

3️⃣ Install Dependencies

pip install -r backend/requirements.txt

4️⃣ Configure Environment Variables

Create a .env file in the project root:

# MongoDB
MONGO_URI=mongodb://localhost:27017
MONGO_DB=insideX
MONGO_COLLECTION=news

# API Keys
perplexity_api=your_perplexity_api_key
perplexity_url=https://api.perplexity.ai/chat/completions
ANTHROPIC_API_KEY=your_anthropic_api_key

5️⃣ Start Server

uvicorn backend.main:app --reload

---

🔎 Usage

API Endpoint: Analyze Stock

POST /analyze Request body:

{
  "stock": "Tata Motors"
}

Response:

{
  "stock": "Tata Motors",
  "perplexity_analysis": "...",
  "retrieved_context": [...],
  "final_report": [
    "• Anomaly detected in EV unit... (source link)",
    "• No issue found in quarterly earnings... (source link)",
    "• Threat Score: 7/10"
  ]
}

---

📊 Example Workflow

1. User queries: "Analyze Tata Motors"
2. Perplexity fetches latest stock news.
3. News gets stored in MongoDB + Pathway memory (for RAG).
4. Claude AI summarizes anomalies into 5 bullet points.
5. A Threat Score (0–10) is generated based on past similar cases.
6. User can ask follow-up questions (“What was all-time high?”).

---

🌱 Future Work

• 📈 Add frontend dashboard (React + Tailwind)
• 🔔 Real-time alerts (email/Slack integration)
• ⚡ Upgrade Pathway to ANN index (HNSW/USearch) for scaling
• 📊 Add stock price + chart integration
• 🔮 More advanced risk modeling

---

🤝 Contributing

Contributions, issues, and feature requests are welcome! Open a PR or start a discussion under Issues.

---

📜 License

MIT License © 2025 insideX