ThutoMo35/African-Stock-Exchange

# African Stock Exchange AI — Ensemble Intelligence System An artificially intelligent, multi-model **ensemble** system for signal generation and research across African equity markets (JSE, NGX, NSE Kenya, EGX, GSE, BRVM, and others). ## Honest scope note This project is deliberately **not** marketed as a "500 billion parameter" model. That figure does not correspond to anything meaningful for equity time-series prediction: - No African-market dataset (public or private) has remotely enough tokens/rows to justify a model of that scale — you would overfit catastrophically. - 500B parameters is GPT-4-class LLM territory, built for language, not tabular/time-series finance data. - What actually improves predictive robustness in quant finance is **ensembling diverse, independently-erroring model families** — not raw parameter count in a single network. So instead, this repo implements a genuine **ensemble AI/ML system**: several structurally different models, each contributing an uncorrelated view, combined by a trained meta-learner (stacking). That's the architecture that's actually used in serious quant research, and it's the one built here. **This is a research/engineering scaffold, not investment advice, and is not connected to a live brokerage or exchange feed.** Wire in real data feeds and paper-trade extensively before ever considering live capital. ## Architecture For the full system blueprint (layer-by-layer breakdown, extension points, and the plan → execute → report → review loop the automation runs on), see **ARCHITECTURE.md**. For the phased roadmap and daily target definition, see **TODO.md**. ``` ┌─────────────────────────┐ │ Feature Engineering │ │ (technical + statistical)│ └─────────────┬────────────┘ │ ┌───────────────┬──────────────┼──────────────┬───────────────┐ ▼ ▼ ▼ ▼ ┌───────────────┐ ┌─────────────┐ ┌───────────┐ ┌────────────────┐ │ Gradient Boost │ │ LSTM Seq. │ │Transformer│ │ Stati …