Loading...
Backend Dev
ML Pipeline Architecture
Featuring:
Multiple Domains, Multiple Signals
Automated Weekly Forecasting
Unified Feature Engineering
6 Parallel GPU Workers
Cross-Domain Covariates
Continuous & Binary Targets
Dual Cloud Storage
Accessibility Technology
Systems Engineering
Data Mining
Forecasting
Enabled Autonomy
Accessibility Technology for Watching Videos
Featuring:
Automated Browsing Experience
Simple Touch for Selection
Responsive Visual & Audio Feedback
Auto Screen Off if Unused
Disabled System Menu and Display
Cloud Synced Videos (No Download Delays)
An accessibility-focused video player. This app enables autonomy for individuals with cognitive impairment to browse and enjoy videos independently using a tablet.
| Forecast Weeks N=54 | Running Avg | % Diff | 3-Week Avg | % Diff | 2026-08-30 Last Week | 2026-09-06 Current Week | |
|---|---|---|---|---|---|---|---|
| Weekly Usage Hours | SD: 13.0 | 24.3 | .36 | 33.0 | .13 | 37.4 | |
| Baseline Forecasts | 22.9 | 1.02 | 46.2 | -.01 | 45.8 | 40.5 | |
| Enhanced Forecasts | 23.1 | 1.08 | 48.0 | -.07 | 44.7 | 38.1 |
Many behaviors fluctuate in response to weather changes and seasonal trends. As a bonus, expert weather forecasts are readily available for usage or benchmarking forecasting methods.
| Forecast Weeks N=54 | Running Avg | % Diff | 3-Week Avg | % Diff | 2026-08-30 Last Week | 2026-09-06 Current Week | |
|---|---|---|---|---|---|---|---|
| Local Max Temp (F) | SD: 18.8 | 68.9 | .24 | 85.3 | .15 | 98.1 | |
| Baseline Forecasts | 68.8 | .28 | 87.9 | -.06 | 82.7 | 96.4 | |
| Enhanced Forecasts |
Economic indicators: VIX, rates, inflation, and more
| Forecast Weeks N=53 | Running Avg | % Diff | 3-Week Avg | % Diff | 2026-08-30 Last Week | 2026-09-06 Current Week | |
|---|---|---|---|---|---|---|---|
| Federal Funds Rate | SD: 0.19 | 3.75 | -.03 | 3.63 | -.00 | 3.63 | |
| Baseline Forecasts | 3.76 | -.03 | 3.63 | .00 | 3.63 | 3.63 | |
| Enhanced Forecasts |
Leveraging a weekly forecasting framework, this project provides predictions for upcoming stock values and buy/sell recommendations.
| Forecast Weeks N=54 | Running Avg | % Diff | 3-Week Avg | % Diff | 2026-08-30 Last Week | 2026-09-06 Current Week | |
|---|---|---|---|---|---|---|---|
| Dow Jones ETF (DIA) | SD: 26.6 | 484.0 | .10 | 534.5 | -.00 | 534.1 | |
| Baseline Forecasts | 483.1 | .11 | 536.4 | -.00 | 534.9 | 534.1 | |
| Enhanced Forecasts | 483.3 | .11 | 535.4 | .00 | 535.7 | 535.6 |
You are here. This web app is designed to showcase the evolution of recommendation-focused data projects, supporting insights with refreshed data, ongoing forecasts, and documented progress. [Code Repo]
| Forecast Weeks N= | Running Avg | % Diff | 3-Week Avg | % Diff | Last Week | Current Week | |
|---|---|---|---|---|---|---|---|
| Signal | SD: | ||||||
| Baseline Forecasts | |||||||
| Enhanced Forecasts |
This backend system provides ongoing data integration and machine learning model management, supporting this web app with curated data for visualizations and data-driven recommendations.
| Forecast Weeks N= | Running Avg | % Diff | 3-Week Avg | % Diff | Last Week | Current Week | |
|---|---|---|---|---|---|---|---|
| Signal | SD: | ||||||
| Baseline Forecasts | |||||||
| Enhanced Forecasts |
Search term trends (e.g., from trends.google.com) can gauge levels of interest from the general population. These signals are useful across many domains, such as understanding demand for various products or activities.
| Forecast Weeks N=54 | Running Avg | % Diff | 3-Week Avg | % Diff | 2026-08-30 Last Week | 2026-09-06 Current Week | |
|---|---|---|---|---|---|---|---|
| Search: 'coupons' | SD: 6.84 | 67.2 | .05 | 70.7 | -.08 | 65.0 | |
| Baseline Forecasts | 68.0 | -.05 | 64.8 | .01 | 65.3 | 66.0 | |
| Enhanced Forecasts |
Weekly availability of foundational AI providers (OpenAI, Anthropic, Google Cloud AI). Outage data from each provider's status page; tracked as a candidate covariate for tech-equity forecasting models.
| Forecast Weeks N=4 | Running Avg | % Diff | 3-Week Avg | % Diff | 2026-04-20 Last Week | 2026-04-27 Current Week | |
|---|---|---|---|---|---|---|---|
| OpenAI Availability (%) | SD: 4.78 | 92.8 | -.03 | 90.4 | |||
| Baseline Forecasts | 98.6 | -.12 | 86.8 | 85.4 | |||
| Enhanced Forecasts |
Strategy Record Review
The comb-through built on what persists, not on a leaderboard
What Persists Per Stock
Alpha, hit rate and skill do not; volatility and drawdown do
The Persistence Control
Skill correlation across halves +0.76; three of four signals positive twice
/strategy-record and Dashboard Header Rework
Forward record, vol tiers, selection monitor; nav entry; rotated signal rail
The Persistent Core
10 terms positive in both halves of the record; four above 5% twice
Measure the Dashboard, Don't Guess
273 mounted Plotly charts were the cost, not the bytes
The Refit Boundary
One training cohort, one decision rule, fresh data verified first
Evidence Admission Policy
Pool breadth costs skill; admit on non-inferiority, escalate on use
Gold and Precious-Metals Complex
26 screened, 8 admitted, duplicates rejected
volgate_v1: Gate on Vol, Not Direction
Forecasts capture ~0% of directional edge; vol timing beats random de-risking by 3 sd
Probabilities Are Informationless
A constant base rate beats them; the moving threshold was retired
Macro Is an Evidence Domain
Own skill weak; as evidence inside stock models it earns its keep
MACRO2: Variance Risk Premium and SKEW
Strategy-aligned additions, derived free from data already on disk
SQL Inspection Layer
A DuckDB warehouse over every artifact, with standing invariant tests
Search Demand Is the Product
26 of 82 terms beat the random walk; the top-10 most forecastable signals are all here
Bit-Deterministic, Shrunk, Append-Only
The run-noise root cause, and a universal accuracy gain
Shrinkage Confirmed on the Control
+2.2 points held-out; the domain that wanted the least shrinkage
Pipeline Validated on Known Ground Truth
SYNTH1: learns real structure, shows no leak on a random walk
22-Year Long-History Base
82 terms stitched 2004 to present, one frozen vintage: +10.8 points of skill
Two-Tier Strategy Measurement
Iterate fast in the lab, prove slowly in an append-only record
Usage Hours as a Forecast Target
Beats persistence; excluded as evidence for everyone else
riskparity_v1 Paper-Trades Forward
Inverse-vol sizing with no signals, frozen and settling weekly
/methods and /features
The experiment programme and feature lab, powered by the warehouse
Enhanced Trends and the Preis Rule, Closed
Covariate evidence adds no skill; 'debt' has no edge on today's data
Client-Legibility Release
Registry-driven names, 165 signals, three-layer publish
Tier-2 Roster and Honest Uncertainty
16 to 60 stocks, ±1σ trailing-vol bands on every chart
Wave-2 Triad Closed in a Day
Direction, volatility, cross-section, horizon: six pre-registered A/Bs
Dashboard Index/Detail Split
The page had died at 60 stocks; heap 1.5 GB to 99 MB
Tier-2 Terms
30 to 82 search-demand signals, each matched to a stock where one exists
Delta Targets, and the Leak They Exposed
94.9% direction on SPY was the alarm, not the discovery
Forecast Skill Panel
The capability claim, with its yardstick
Weather as the Control Domain
Falsifying the machinery on purpose: 3 of 4 signals beat every naive rule
Direction-Skill Audit
5,088 graded calls, no slice with skill, confidence anti-informative
Blinded Validation Caught a 10x Overstatement
Unblinded said +10.5% alpha; blinded said +1.79%
Sunday Fetch Resilience
VPN rotations were tearing the host network mid-fetch
Onboarding Hub
A dated front door: docs map, glossary hovers, live state
Google Rewrites Its Own History
Renormalization measured in the stored fetches; the read-path fix rejected by its own gate
/review: Replay Any Sunday Run
Fast-forward or step-by-step, with the verification checklist as code
TV Series on Their Own Rail
Engagement promotes the next episode rather than protecting the current one
Dev / Prod Split
Inert dev container, prod gated behind an explicit flag
Engagement-Based Weekly Video Shuffler
Rollout week 0 (rollout: 2026-05-24)
Lifecycle Measurement Layer
Stints, strikes and evergreens: grading every pick by what it earned
Engineering Handbook
Numbered chapters, onboarding tour, communication doctrine
Paper-Trading KPI Layer
Risk metrics, portfolio aggregator, 61-week binary backfill
The Honest KPI
53-week walk-forward: not beating buy-and-hold
Kiosk Weekly Refresh
The locked tablet picks up Sunday's shuffle without a redeploy
Macro Domain Onboarded
18 indicators: rates, spreads, FX, commodities, valuation, volatility
A/B Validation Harness
Seeded determinism, cost sweep, snapshot lineage
UI Enhancements
Stats Showcase (Animation & Annotation)
UI Enhancements
Replacing RMSE with 5-Star system
Dynamic Dashboard Pages
Links to Dashboards from Project Cards
Correcting Sunday Week Start
Accounting for Time Zone Shifts
Blob Storage for Dashboard Data
Serve Curated Data Files Across Projects
Scheduled Data Processing
Automation for Weekly Processing
v1 Baseline & Enhanced Modeling
Google Trends Data Helps Predict Usage
Baseline Weather Forecasts
Weekly Forecasts Using Prior Signal Data
Following Closing Price Values
Outline Scope and Success Markers
Featured Weekly Plot (plotly.js)
Wider View of Featured Data
V1 Weather Signals (Local)
Source Data and Processing
Novel and Sparse Time-Series Features
Feature Engineering Time and Domain Data
GPU and Parallel Compute (CUDA image)
Reconfiguring Backend System for CUDA
Featured Data SVG Component
Time Series Lines and 'Last Week' Points
Keywords for Predicting Video Watching
Air Conditioner, Indoor, Outdoor, Showtimes
Web Search Trends for Prediction
Methodology from Preis, Moat, and Stanley (2013)
v2 Video App for Cognitively Impaired
Quality of Life Changes
v1 Video App for Cognitively Impaired
Early Testing and Observation
v1 Video App for Cognitively Impaired
Data Fetching and Processing
v1 Data Warehouse
Fetch and Processing System
Data Warehouse Development
Outline Scope and Success Markers
v1 Video App for Cognitively Impaired
Data Processing -- Usage Hours
v1 Video App for Cognitively Impaired
Backend Service for Data Processing
v1 Projects Web App
Project and Progress Components
v1 Projects Web App
Development Page Layout
Tracking US Metro Location Housing Values
Foundational Data
v1 Projects Web App
Outline Scope and Success Markers
v1 Video App for Cognitively Impaired
Touch Tablet Selection, Passive Browsing
v1 Video App for Cognitively Impaired
Outline Scope and Success Markers

Hi, welcome.This website provides weekly reporting for the Enabled Autonomy Video Player tablet and other data signals. The main goal is to support evidence-based improvements for the app and other projects by monitoring and forecasting key measurements.I created the Enabled Autonomy Video Player app for my cousin, Bill (cognitively impaired), to provide him a video browsing and viewing experience he can engage with independently.

Data Scientist
2021 - 2024
Standardized data pipelines for order, customer, and store-level measurement, supporting business-wide analytics (marketing promotions, test items, operational changes).
Created customer segmentation methodology based on retention/churn modeling, improving ad spend efficiency.
Led development for item-specific demand predictions served throughout the day to store operators at each domestic and international location.







Data Scientist
2019
Designed a data pipeline and Bayesian regression model optimizing channel-by-funnel ad spending.
Developed a dashboard visualizing ad-spend response curves for channel-by-funnel targets.
Reviewed existing advertising testing methods, including geospatial A/B testing.







Postdoctoral Researcher and Course Instructor
2014 - 2018
Employed data science methods for memory-based decision-making research.
Designed novel computer-based experiments and evaluated models with behavioral data.
Secured a 2-year postdoctoral fellowship from the Swiss National Science Foundation.




Ph.D. in Cognitive and Brain Sciences
State University of New York at Binghamton
B.S. (Summa cum laude) in Psychology
Eastern Michigan University