SELECTED WORK

Problems solved.
Software shipped.

A sample of tools built for real organizations. Each one was designed to eliminate a manual process, surface better information, or give a team back hours they should not have been spending in the first place.

01 / PROJECT
Proximity Dispatch Tool
CLIENT: NURSE STAFFING & HOME CARE COMPANY
Workforce Operations Proximity Matching Scheduling & Dispatch Benefits Eligibility Retention Home Care

Smart, location aware workforce operations for home care and nursing staffing, unifying caregiver dispatch, onboarding, retention, and benefits eligibility on top of a live sync with the agency's scheduling system. Instead of manually cross referencing who is available and who is nearby, schedulers select an open shift, a client, or a caregiver and instantly see the best matches ranked by travel time, complete with availability, weekly hours, and preferred or exclusion status. What used to take phone calls and spreadsheets now takes seconds.

WHAT IT DOES
Proximity matching, both directions. Pick a shift or client to find the closest available caregivers, or pick a caregiver to surface the nearest open shifts and client locations.
Real travel times. Driving distance and estimated public transit time for every match, plus one click directions.
Double booking and overtime protection. Flags caregivers already scheduled in an overlapping window, and shows weekly hours at a glance with clear alerts before anyone crosses 40 hours.
Applicant proximity. For recruiting, see which clients are nearest to a new applicant to guide placement conversations.
Onboarding and retention dashboard. New hires are tracked by quarter and shift activity, so stalling caregivers get proactive outreach before they churn.
Healthcare benefits eligibility engine. Automatically tallies each caregiver's monthly hours against the eligibility threshold, surfaces who newly qualifies, and manages enrollment, waivers, and offboarding.
One tap outreach. Call caregivers directly through Microsoft Teams from any result.
IMPACT
By turning a slow, manual matching process into an instant, data driven one, the tool helps the agency fill shifts that would otherwise slip through the cracks, capturing an estimated $13K per week in additional billable revenue. It also reduces travel burden on caregivers, prevents accidental overtime, flags at risk new hires before they leave, and keeps benefits eligibility tracking accurate across the entire roster, giving every client a well matched, nearby care partner.
React SPA Node.js backend Microsoft SSO Mapping API Routing engine Workforce API sync Caching architecture Cloud hosting CI/CD deployments
02 / PROJECT
Teacher-Student Oracle
CLIENT: CAREER COACH
AI Coaching Assistant RAG Semantic Search Source Weighting Coach in the Loop Self Updating Knowledge

A production AI coaching assistant for a career transition program, on call around the clock. It answers in the program's own voice, grounded in course material, the Slack community, and session notes, keeps its own knowledge current, and lets the coach steer it in her own words in real time.

WHAT IT DOES
Always on. Web chat any time. Optional email sign in remembers each student's progress and personalizes as it goes.
Self updating knowledge. Syncs the Slack community weekly, public and private channels, alongside course material and session notes. No manual upkeep, ever.
Trust weighted answers (RAG). Semantic search finds what is relevant; source weighting decides what is trusted, the coach's guidance first, peer chatter lightest. It informs without overriding the teaching.
Coach in the loop. The coach types guidance straight into the Oracle and it becomes top priority instruction instantly, steering a whole cohort without a line of code.
One console. A single sign in to upload content and see exactly who is using the Oracle, how much, and when.
Early warning. Reads the tone of recent messages and flags students who may be struggling, so the coach reaches out before they drift.
Right moment nudges. After sustained use, it points students to the community channel, bridging AI help with real human support.
IMPACT
One coach's expertise, scaled to on brand support at any hour, with no added workload and no content admin. The early warning flag turns support into retention. It all runs on serverless edge infrastructure at low monthly cost, deploying itself on every change.
Anthropic API OpenAI embeddings Cloudflare Workers & Pages Supabase (Postgres + pgvector) Resend email OTP Slack API GitHub Actions CI/CD
03 / PROJECT
Trade Intelligence Platform
CLIENT: MAJOR U.S. MARITIME PORT
Trade Intelligence Data Warehousing Dimensional Modeling Executive Reporting ETL Pipeline Big Data

A modern trade intelligence platform built for one of the most active cargo ports on the U.S. East Coast. The port handles over 36 million tons of cargo annually, reached a record 841,000 TEUs in 2024 (up 13% year over year), and is ranked #1 in North America for container productivity by the World Bank and S&P. This project replaced a fragmented, manual reporting environment with a unified data warehouse, dimensional model, and executive Power BI dashboards that update automatically as new data arrives.

WHAT IT DOES
Entity relationship model designed for scale. A dimensional data architecture built in Snowflake supports a core shipments fact table with 110+ million records, joined to dimension tables for port details, commodity classifications, trade representatives, and shipping lines.
Two ingestion paths. A simple bulk upload path handles files up to 250MB directly in Snowflake. An Azure Blob Storage external stage handles larger files and batch loads, with automated COPY INTO commands staging data without size constraints.
Snowflake views as the reporting layer. Aggregated views join the shipments table to dimension tables, coerce data types, and feed Power BI reports via Direct Query, so dashboards always reflect the latest certified data without manual refreshes.
Multiple report environments. Separate warehouses for data loading and reporting eliminate performance conflicts. A shared Power BI workspace gives approved stakeholders access to East Coast, local, and US trade standard reports without each requiring individual file distribution.
Flexible regional scoping. Dimension tables allow analysts to include or exclude ports from metro or regional groupings at reporting time, without touching the underlying data. Cargo assigned to representatives updates downstream automatically.
IMPACT
A port tracking over $7.3 billion in annual food cargo and nearly 282,000 vehicle imports in 2024 now has a reporting foundation that scales with its 15 year growth plan. Analysts can run live queries across 110+ million shipment records without disrupting end user reporting, and trade representatives see current cargo assignments the moment new certified data is loaded. The underlying trade data was also the foundation of published academic research coauthored at Penn State University, applying ARMA-GARCH, XGBoost, and MLP regression models to analyze the impact of COVID-19 on U.S. maritime trade volumes across 11 years of import and export data.
PUBLISHED RESEARCH
Analyzing U.S. Maritime Trade and COVID-19 Impact Using Machine Learning
Peter R. Abraldes et al.  ·  IGI Global, Big Data as a Service Section  ·  2023  ·  DOI: 10.4018/978-1-7998-9220-5.ch021
Snowflake Azure Blob Storage Power BI Direct Query SQL views PIERS trade data ERD modeling
04 / PROJECT
Field Operations Platform
CLIENT: PRESSURE WASHING CONTRACTOR
Field Operations Crew Management Time Tracking Quoting Photo Documentation South Jersey

A purpose built field operations tool for a fully insured, licensed pressure washing and exterior cleaning contractor serving residential, commercial, and industrial clients. Replaces manual timesheets, handwritten quotes, and unorganized job photos with a single mobile first system crews can use on site.

WHAT IT DOES
Clock in and out from the field. Employees record start and end times from any device, eliminating paper timesheets and ensuring accurate payroll data without a back office reconciliation step.
Quote finalization on site. Crews can review and lock in job details and pricing before leaving a property, reducing follow up calls and keeping estimates tied to actual conditions.
Before and after photo documentation. Structured photo capture at job start and completion creates a timestamped record for every job, supporting quality assurance, dispute resolution, and client communication.
IMPACT
Gives a lean field crew the documentation and operational discipline of a much larger company, with zero extra back office burden. Every job leaves a complete, timestamped record: who was there, what was agreed, and what the work looked like before and after.
Mobile first web app Time clock Photo capture Quote management Cloud storage
05 / PROJECT
Statistical Expert Report
CLIENT: CONSTITUTIONAL LAW FIRM
Applied Statistics Expert Testimony Survey Methodology Sampling Theory Litigation Support

A signed statistical expert report prepared for litigation support, analyzing whether a defendant's collection of 77 affidavits constituted a valid survey capable of being generalized to the broader population of host representatives. The report identified fundamental flaws in experimental design and statistical methodology, quantified the gap between the submitted sample and a properly constructed one, and proposed the correct survey design parameters had a rigorous approach been taken.

KEY FINDINGS
Selection bias invalidated the sample. Affiants were not selected through random sampling, introducing significant selection bias and undermining the ability to generalize findings to the population of approximately 588 host representatives who held 3,764 events during the relevant period.
Effective sample was far smaller than claimed. Nearly 30% of affidavits cited events outside the relevant date range, and another 13% did not cite event dates at all. The usable affiant count was reduced from 77 to 44 after applying date range exclusions.
Respondents were not independent actors. Many affiants had affiliations with the defendant or with one another, violating basic principles of unbiased response collection and compounding the selection bias problem.
Survey instrument lacked methodological rigor. The three affidavit forms lacked the structure, neutrality, nonresponse inclusion, and anonymization required of a valid survey. Language relied on stated belief rather than demonstrated understanding and did not control for confounding variables.
Required sample size calculated and documented. Using a Chi-squared proportional means test framework with 95% confidence, 80% power, and a finite population correction for the 588 person population, the minimum valid sample was calculated at 468 respondents under equal proportions: ten times the effective affiant group.
DELIVERABLE
An 18 page signed expert report submitted under professional certification, covering population definition, distribution analysis, date range exclusions, sample independence, instrument quality assessment, Chi-squared test design, finite population correction methodology, and a scenario for calculating an appropriate sample size. Conclusions held to a reasonable degree of professional certainty.
Chi-squared test Finite population correction Power analysis Sample size calculation Selection bias audit Signed expert report
Have a problem worth solving?

These started as real operational headaches. If your team is losing time to something that should be automated, tracked, or just better, reach out and let us talk about it.

Start a conversation