V

What I've Built

9 projects shown

A 24 hour ERCOT demand forecast with its 80 percent interval shaded behind the forecast median, plotted against the demand that actually occurred, with the forecast origin marked

Bellwether

A dashboard over two years of hourly US electricity demand from EIA Open Data, forecasting the next 24 hours for ERCOT, CAISO, and PacifiCorp East and scoring every interval against what actually happened.

PythonChronos-BoltTimesFMDuckDB
Propensity score overlap histogram comparing treated and control patients, with common support cutoffs marked at 0.05 and 0.95

RHC Causal Inference

Estimates the causal effect of right heart catheterization on ICU mortality using propensity score weighting, doubly robust estimation, and E-value analysis.

PythonCausal InferencePropensity ScoresIPW
Readmission risk predictor interface showing a patient's predicted probability next to a SHAP waterfall of the features driving it

CMS Readmission Predictor

Predicts 30-day unplanned hospital readmissions using Medicare claims data. LightGBM classifier with monotonic constraints for clinical correctness, SHAP interpretability, and full experiment tracking via MLflow.

PythonLightGBMDuckDBSHAP
Architecture of the conversational intelligence platform: five upstream systems consolidating into a canonical interactions layer, batch LLM enrichment, structure-aware chunking, hybrid retrieval, and a ReAct router serving a chat and filter interface

Conversational Intelligence

A retrieval system over 225,000+ customer conversations that answers questions in the words customers actually used, with citations back to the specific interaction.

BigQueryVertex AIVector SearchCloud Run
Diagram of the REDX email pipeline: customer data platform feeding a segmentation rules engine, mutually exclusive audience segments, the campaign engine, and outcome tracking

REDX Email Operations

The email infrastructure at REDX: roughly 800K sends a month to 100K+ contacts, across a rules engine that produces mutually exclusive audience segments.

Customer Data PlatformSales PlatformData Visualization
Executive reporting dashboard showing lead totals, MQL and SQL counts, lead velocity over 30 days, lead source mix, email engagement, and an SEO snapshot

Executive Reporting Pipeline

The GCP data platform behind marketing reporting: BigQuery warehouse, scheduled transformation jobs, a curated core table layer, and reverse ETL back into the operational tools.

BigQueryGCPCloud FunctionsCloud Storage
Diagram of the lead pipeline: capture sources flowing into identity resolution, enrichment, and MQL qualification in the customer data platform, then routing and SDR outreach in the sales CRM

Lead Data Pipeline

Carries a lead from capture into the sales platform without anyone touching it: identity resolution, enrichment, and qualification, at a one-minute median.

Customer Data PlatformSales PlatformAPI IntegrationIdentity Resolution
The Expense Snap app icon, a green lightning bolt monogram on a white tile, set on a dark background

Expense Snap

Android and iOS expense tracker that reads a receipt on the device and writes it to a Google Sheet you own. No server, no cloud OCR, and no third party ever sees the data.

FlutterDartGoogle ML KitOn-Device OCR
Diagram of the Atreus orchestration system: an orchestrator directing six specialist agents for planning, building, validation, scheduling, and archiving, with persistent memory and verification gates

Atreus

A self-hosted agent system I describe work to in plain language: it branches, edits, builds, verifies the output against the real artifact, and reports back with a preview link.

PythonDockerAI OrchestrationVector Database