An advanced, AI-driven quantitative prediction and interpretation system designed specifically for Indian financial investors and beginners. Powered by Kronos-base (the ~400MB open-source candlestick foundation model trained on global exchanges) and localized by Llama 3.2 3B via Ollama, this project translates complex K-line sequences into jargon-free, high-value financial insights for equities like NIFTY 50, Reliance Industries, HDFC Bank, Gold ETFs, and USD/INR exchange rates.
- Kronos Foundation Model: Utilizes the 102.3M parameter (
~400MB) base model with hierarchical K-line tokenization to predict future OHLCV candlestick paths. - Beginner-Friendly Interpretation: Local Ollama AI analyzes quantitative forecasting metrics and translates them into intuitive analogies tailored to Indian market participants.
- Empirical Proof Testing: Includes an automated quantitative validation script to measure directional accuracy and support/resistance resilience against historical NSE data.
- Responsive HTML Visualizers: Generates aesthetic, interactive reporting templates featuring dark-mode glassmorphism and instant markdown export capabilities.
- Quantitative Forecaster:
NeoQuasar/Kronos-base(Hugging Face Transformers & PyTorch) - Local Explainer AI:
llama3.2:3b(via Ollama, thinking disabled for rapid inference) - Market Data Provider:
yfinance(Live NSE Indian market indices, corporate stocks, and commodities) - Visual Templating:
Jinja2with custom responsive CSS styling - Data & Math Engine:
pandas,numpy,matplotlib,torch
All commands below can be executed directly in your Windows Antigravity PowerShell terminal by copy-pasting.
First, clone the official Kronos foundation model repository into kronos_lib and install standard Python dependencies:
git clone https://github.com/shiyu-coder/Kronos.git kronos_lib
pip install --upgrade transformers huggingface_hub safetensors pandas einops matplotlib tqdm yfinance jinja2 requests torchEnsure Ollama is running, and pull the required 3B instruction model:
ollama run llama3.2:3bThe first execution automatically caches the ~400MB NeoQuasar/Kronos-base model and NeoQuasar/Kronos-Tokenizer-base from Hugging Face Hub into your system storage.
Run the primary pipeline to fetch latest Indian market K-lines, generate 14-day predictions, compute support/resistance levels, and produce beginner-friendly advisory reports:
python market_advisor.pyUpon execution, the system creates:
outputs.md: A structured markdown executive summary of forecasts.indian_market_report.html: A visually immersive HTML report with download triggers.
Validate Kronos-base predictive directional reliability by backtesting recent historical periods against actual ground truth closing prices on Indian symbols:
python proof_test.py| File / Folder | Description |
|---|---|
market_advisor.py |
Main application script orchestrating data retrieval, Kronos K-line forecasting, LLM translation, and report output generation. |
proof_test.py |
Supporting backtest verification tool computing precision and directional win-rate metrics on NSE symbols. |
templates/report_template.html |
Jinja2 responsive HTML visualizer template featuring dynamic UI styling and standalone export features. |
outputs.md |
Automated generated advisory output detailing target levels and simplified Indian financial commentary. |
kronos_lib/ |
Core library folder cloned from official Kronos repository providing autoregressive Transformer and hierarchical tokenization architectures. |
- Retail NIFTY 50 Trend Navigation: Helps novice Indian investors understand market momentum without mastering complex technical indicators like RSI or MACD.
- Blue-Chip Swing Level Identification: Automates the detection of entry and profit-booking targets for high-liquidity Indian Equities (e.g., Reliance Industries, HDFC Bank).
- Gold ETF & Rupee Volatility Hedging: Monitors commodities and USD/INR exchange paths to guide risk mitigation for household portfolios and importers.
- Jargon-Free Financial Literacy: Converts complex quant output (standard deviation, nucleus sampling intervals, trend lines) into simple explanations using everyday analogies.
- Pre-Market quantitative Screening: Enables disciplined traders to objectively benchmark their qualitative trading hypothesis against an unbiased foundation model.
- Multi-Timeframe Confluence Engine: Combining 5-minute intraday candlesticks with daily closing sequences to generate intraday scalping insights.
- Indian Economic Events Integration: Injecting RBI monetary policy schedules and quarterly corporate earnings dates directly into Ollama context windows.
- Interactive Charting with Lightweight Charts: Transitioning standard visual generation to full real-time interactive HTML5 trading canvases.
- Automated Telegram Alert Bot: Periodic scheduled cron execution delivering customized morning advisory notes directly to mobile messaging channels.
- Custom NSE Finetuned LoRA Weights: Specializing the core Kronos foundation architecture directly on 15-year micro-cap and small-cap Indian market sequences.
kronos-ai time-series-forecasting candlestick-prediction nifty50-forecasting indian-stock-market financial-foundation-model ollama llama3.2 algorithmic-trading quant-finance python-trading-bot
Built with simplicity, elegance, and AI quantitative accuracy for the Indian financial ecosystem.