⚡ AI Quant-Coder

AMD Developer Hackathon: ACT II Submission

An end-to-end Financial Market Predictor and Retrieval-Augmented Generation (RAG) Coding Assistant powered by open-source LLMs on AMD Developer Cloud GPUs.

AI Quant-Coder bridges the gap between financial time-series forecasts and algorithmic trading script generation — combining a statistical prediction engine with a full in-browser coding assistant that generates, explains, and structures quantitative trading bots in real-time.

---

🖥️ Live Application Pages

Page	Description
📈 Market Predictor	Live stock charts with technical indicators (SMA, EMA, RSI) and 30-day ARIMA price forecasts. Includes an embedded code assistant chat.
💬 AI Code Assistant	Full-screen Gemini-style coding assistant with multi-session chat history, hamburger sidebar toggle, and RAG DB manager.
⚙️ System Configuration	Configure local Ollama or AMD Developer Cloud vLLM API endpoints and model identifiers. Run live connection diagnostics.
ℹ️ About Project	Detailed project architecture cards, module deep dives, and core hackathon innovation feature highlights.

---

🚀 Key Features

📈 Market Predictor Engine

• Fetches up to 10 years of historical stock data from Yahoo Finance API
• Computes rolling technical indicators:
  • SMA 20 / SMA 50 — Moving Average Crossover detection
  • EMA 20 / EMA 50 — Exponential Moving Averages
  • RSI (14-day) — Relative Strength Index (Overbought/Oversold)
• ARIMA(1,1,0) Forecasting — Statistical 30-day close price projection
• Interactive Plotly Charts — Dark-mode, zoom-enabled candlestick/line layouts
• Swappable Panel Layouts — Swap graph and chat positions inline on the page

💬 RAG Coding Assistant

• TF-IDF Vector Database — Indexes local open-source trading script templates
• Semantic Code Retrieval — Finds relevant Python code templates using cosine similarity matching
• Predictor Context Injection — Automatically appends live RSI, SMA trend, current close, and 30-day ARIMA forecast into every LLM request
• Multi-Session Chat History — Persistent JSON-backed session store with named conversations
• Thread-Isolated Background Streaming — LLM generation runs on background worker threads; UI never freezes
• Active Session Indicator — Active chat highlighted with bright green gradient and 🟢 marker

🔌 AMD ROCm / vLLM Ready

• Configurable API endpoint routing for:
  • Local Ollama — http://localhost:11434/v1
  • AMD Developer Cloud vLLM — Any remote OpenAI-compatible endpoint
• Live connection diagnostic panel to verify endpoint/model health

---

---

⚙️ Local Development Setup

1. Clone the Repository

git clone https://github.com/<your-username>/Hack-tn.git
cd Hack-tn

2. Create Virtual Environment & Install Dependencies

# Create environment
python -m venv venv

# Activate — Windows
.\venv\Scripts\activate

# Activate — Linux / macOS
source venv/bin/activate

# Install all dependencies
pip install -r requirements.txt

3. Set Up LLM Backend

Option A: Local Ollama (Recommended for quick testing)

# Install Ollama from https://ollama.com
ollama pull llama3.2

# Ollama auto-serves at: http://localhost:11434/v1

Option B: AMD Developer Cloud (vLLM + ROCm)

# On your AMD GPU Ubuntu server:
source venv/bin/activate
pip install vllm  # ROCm-compatible build

# Serve model with vLLM
python -m vllm.entrypoints.openai.api_server \
    --model meta-llama/Meta-Llama-3-8B-Instruct \
    --port 8000

# Update the endpoint in the app: http://<server-ip>:8000/v1

4. Run the Application

streamlit run app.py

Opens at http://localhost:8501 in your browser.

---

🧠 How the RAG Pipeline Works

User Prompt
    │
    ▼
TF-IDF Search (rag_module.py)
    │  Finds top-k matching code template chunks
    ▼
Predictor Context Compiler
    │  Appends live stock stats: RSI, SMA trend, close price, ARIMA forecast
    ▼
Augmented Prompt Builder
    │  Merges code templates + financial data into structured LLM instruction
    ▼
LLM Streaming Client (llm_client.py)
    │  Sends augmented prompt via background thread
    ▼
Script Output (Streamed to chat in real-time)

---

📄 License

This project is submitted as an open-source hackathon entry. Free to use and extend for educational and research purposes.

---

👥 Authors

Name	Role
Agha Essa Khan	Team Leader & AI Architect
Misbah Ramzan	Quantitative & Python Developer

Submitted for the AMD Developer Hackathon: ACT II