BlueX — Investment Intelligence for Retail Investors
A systematic momentum-scoring and portfolio-research platform that turns multi-timeframe signals into evidence-led decisions for portfolio management.
MY ROLE
End-to-end analyst and builder: indicator design, CSV pipeline, validation, ML experiments, and the public intelligence-board interface.
THE DECISION
Which assets show the strongest current momentum, and is that signal statistically meaningful enough to inform portfolio weighting?
APPROACH
- Built a single canonical producer for multi-timeframe feature CSVs
- Validated momentum families with ANOVA-style checks on decision labels
- Ran tree/sklearn research and walk-forward checks on Mac; DL on RunPod only
- Published a scrubbed public package: renamed light codes, remodel summaries, and UI evidence
EVIDENCE & DELIVERABLES
- Current HUD stats/subplot/ANOVA pilot using public L#### / G## codes
- Prior ML remodel summaries: ensemble, XGBoost, LightGBM, and MLP
- 18 scrubbed daily ensemble dashboards with Plotly drill-down
- Core research codebase remains proprietary; public case shows workflow and UI only
SELECTED WORK FILES
Research Overview
Sanitized entry point for current HUD links, remodel summaries, model compare, and capital-set sample.
4 KBCurrent Signal Overview
Current lights layer with public codes only — stats, NFLX subplot, and ANOVA pilot.
5 KBEnsemble Summary
Strategy return, alpha vs buy-and-hold, Sharpe, and ML accuracy for the ensemble remodel.
10 KBXGBoost Summary
Gradient-boosted remodel metrics on the same holdout framing as the other tree models.
16 KBLightGBM Summary
LightGBM remodel metrics compared against ensemble and XGBoost on identical splits.
16 KBMLP Summary
MLP remodel summary table for the same ticker universe.
15 KBModel Comparison
Side-by-side remodel comparison without private feature identifiers.
11 KBDecision Dashboard
Decision-focused public view; formulas and private recipes stay private.
13 KBEVIDENCE INTEGRITY
This case is presented as proprietary research system · public pages are scrubbed metrics/ui only · not trading advice. The label distinguishes observed work from simulated impact, proxies, historical comparisons, or proposed architecture.