End‑to‑end trading analytics built on open/public data feeds (prices + news), explainable ML (XGBoost + SHAP), and serverless inference on AWS. UI is a React app (no Streamlit).
Core idea: ingest fresh market + news signals, build features, serve real‑time predictions with plain‑English explanations, and keep everything cheap, observable, and auditable.
- Amazon S3 — raw data, features parquet, model artifacts
- Amazon DynamoDB — per‑day records (
prediction,news,shap,explanation) - AWS Lambda — price/news collectors, feature builder, refresh hook
- Amazon EventBridge — schedules for polling/ETL cadence
- Amazon SageMaker Serverless — low‑cost, hot model inference
- Amazon SageMaker (Training) — notebooks/scripts to train model.pkl
- Amazon Bedrock — converts SHAP + metrics to short explanations
- Amazon API Gateway — public API surface for UI
- Amazon CloudWatch — logs, metrics, alarms
- Backend/ML: Python, XGBoost/LightGBM, Pandas, PyArrow, SHAP
- Infra: AWS CDK or Terraform (IaC), IAM least privilege
- UI: React (Vite or Next.js), TypeScript, TanStack Query, Tailwind (optional)
- Data: Stooq (prices), GDELT (news). Optional Finnhub/Polygon/Schwab for inference
- Testing: Pytest, Jest/Vitest
- CI/CD: GitHub Actions (lint, unit tests, CDK/Terraform plan & deploy)