Projects

Portfolio work built on synthetic or public data. Client and employer projects remain private.

Hospitality Analytics Subject: urban hotel · Comp set: 4 properties, simulated 119,390 bookings, 2015 to 2017 · July 2026

Hotel RevPAR Performance Readout

The hotel charges more than its comp set and fills fewer rooms.

A monthly STR-style performance report for an urban hotel: occupancy, ADR, RevPAR, and market share indices against a competitive set. Built from 119,000 real bookings through a bronze, silver, and gold Delta pipeline on Databricks.

Exhibit 1. Monthly readout, subject vs comp setSix months
STAR-style summary table: monthly occupancy, ADR, and RevPAR for the subject property and comp set, with MPI, ARI, and RGI index rows.

Source: Kaggle Hotel Booking Demand dataset. Comp set simulated from the source data.

103.5
RevPAR index (RGI), trailing 12 months
Above comp set
106.1
ADR index (ARI)
Above comp set
97.5
Occupancy index (MPI)
Below comp set
119K
Bookings processed
19 automated tests

Recommendation

To the General Manager

Hold rate. The property earns more per available room than its comp set because its rate premium outweighs an occupancy shortfall. The shortfall is concentrated in midweek, short-lead corporate business that a discounting competitor is winning. Address it with targeted corporate rates and a two-week booking offer rather than a general rate cut.

A general rate cut would give up revenue the property already earns to buy occupancy it can win at lower cost.

Results

The property's RevPAR index is above 100 in every month of the trailing twelve. The gain comes from rate: the ADR index is above 100 throughout while the occupancy index is below it.

Exhibit 2. RGI, trailing 12 months
RGI trend over twelve months with the 100 index line marked and share gain shaded.
Exhibit 3. ADR and occupancy vs comp set
Side-by-side ADR and occupancy trends for subject property versus comp set.

Occupancy

The occupancy shortfall is concentrated midweek, where a discounting competitor is taking share. Segment mix and booking pace point to transient corporate demand booked inside two weeks.

Exhibit 4. Segment mix vs comp norm
Revenue mix by market segment, subject property versus simulated comp set norm.
Exhibit 5. Booking pace, 2017 vs 2016
Booking pace curves: share of final room nights on the books by days before arrival, 2017 versus 2016.

Pipeline

Bronze

Raw bookings are loaded to a Unity Catalog volume and written to a bronze Delta table unchanged, with an ingestion timestamp.

Silver

Arrival dates assembled, cancellations typed, zero-guest and negative-rate rows removed, one €5,400 ADR outlier capped, market segments normalized to five values. Data-quality checks stop the run on any failure.

Comp set

Four competitor hotels simulated from the subject's daily history with different pricing behavior. Results are aggregated room-weighted, following STR practice.

Gold

Bookings expanded to stay nights for daily occupancy, ADR, and RevPAR. Monthly rollups add MPI, ARI, and RGI. Booking pace measured at 120, 60, and 30 days before arrival.

Background

Hotel revenue analyst roles require reading STR reports, so this project follows STR conventions: RevPAR index, MPI, ARI, comp set, pace, and pickup. The pipeline is built as a production Databricks lakehouse would be. The output is the monthly report a general manager receives.

Portfolio project on public data. The competitive set is simulated and does not represent a real market. Code and tests are on GitHub.

Data Engineering AAPL, TSLA, AMZN, GOOGL, MSFT · Daily prices, six months AWS and Databricks · April 2026

Stock Market Analytics Pipeline

Getting five stocks from Yahoo Finance into a Delta table, with pattern analysis run through AWS Bedrock.

An end-to-end cloud pipeline: live stock prices ingested to S3, analyzed by Claude through AWS Bedrock, stored in a Delta Lake table on Databricks, and reported with Databricks SQL.

Exhibit 1. Monthly average closing price, five tickersSix months
Monthly stock price trends from the Databricks Delta Lake table.

Source: Yahoo Finance daily closes, resampled monthly in Databricks SQL.

5
Tickers tracked
AAPL, TSLA, AMZN, GOOGL, MSFT
6mo
Daily price history
Yahoo Finance API
4
Pipeline stages
Ingest, analyze, store, report
ACID
Delta Lake storage
Schema enforcement, time travel

Summary

Python pulls six months of daily prices for five large-cap stocks and lands them in S3. A Lambda function triggers on each new file, sends the data to Claude through AWS Bedrock for pattern analysis, and writes the results back to S3. Databricks loads the data into a Delta Lake table and Databricks SQL produces the trend reporting.

Results

Databricks SQL reports monthly average prices, price ranges, best and worst trading days, and year-over-year comparisons for each ticker.

Pipeline

Ingest

Python pulls six months of daily prices for the five tickers from the Yahoo Finance API, cleans and normalizes them, and uploads structured CSV to S3.

Analyze

Lambda triggers on new S3 objects, reads the CSV, sends it to Claude through AWS Bedrock for pattern analysis, and writes the output back to S3.

Store

Data is loaded into Databricks and saved as a Delta Lake table with ACID transactions, schema enforcement, and time travel.

Report

Databricks SQL computes monthly trends, price ranges, best and worst trading days, and year-over-year comparisons, rendered as dashboards.

Background

Stock analytics is a common first portfolio project. This version runs on cloud infrastructure end to end, uses a language model for the analysis layer, and stores data in a lakehouse with ACID transactions rather than a local file.

Portfolio project on public market data. Not investment advice. Code is on GitHub.

Predictive Modeling Synthetic golf ecommerce retailer · 5,000 customers 65,000 transactions · April 2026

Customer Lifetime Value Model

One Tour Enthusiast is worth ten Weekend Golfers.

Probabilistic customer lifetime value for a synthetic golf ecommerce retailer. 65,000 transactions become per-customer 12-month revenue forecasts using RFM features with BG/NBD and Gamma-Gamma models.

Exhibit 1. Predicted CLV by segment
Predicted customer lifetime value by segment: average CLV per customer on the left, total CLV per segment on the right.
$3.1M
Predicted revenue, next 12 months
Summed per customer
10.2x
Top vs bottom segment CLV
Tour Enthusiasts vs Weekend Golfers
70%
Forecast revenue from top two segments
32% of customers
65K
Transactions modeled
5,000 customers, 6 segments

Summary

The top two customer segments produce 70% of forecast revenue from 32% of the customer base, at similar acquisition cost. Tour Enthusiasts have 10.2 times the predicted lifetime value of Weekend Golfers. Predicted next-12-month revenue across all customers is $3.1M.

Results

Predicted lifetime value by segment and for the top ten individual customers, with historical spend shown against the 12-month forecast.

Exhibit 2. Top 10 customers, history vs forecast
Top 10 predicted CLV customers, historical spend vs predicted next 12 months.

Source: synthetic transactions generated for this project. Models: BG/NBD and Gamma-Gamma via the lifetimes library.

Pipeline

Data

A synthetic golf ecommerce dataset generated in Python: 65,000 transactions across 5,000 customers and six behavioral segments, with seasonality, churn, and upgrade cycles.

Bronze

Raw transactions uploaded to S3 and loaded into Databricks as a Delta table with schema enforcement, time travel, and ACID transactions.

Silver

SQL aggregations produce per-customer Recency, Frequency, and Monetary features: one row per customer.

Model

BG/NBD predicts future transaction counts. Gamma-Gamma predicts order value. Combined, they give expected 12-month revenue per customer.

Portfolio project on synthetic data. Segment names and figures are illustrative of method. Code is on GitHub.

Finance Lululemon Athletica (NASDAQ: LULU) · 5-year DCF and reverse DCF SEC EDGAR XBRL, 16 years · May 2026

Lululemon DCF Valuation Pipeline

The market has Lululemon shrinking 9% a year through 2030.

A five-year discounted cash flow valuation of Lululemon built as a data pipeline: SEC EDGAR filings ingested to S3, normalized through bronze, silver, and gold layers, analyzed with SQL window functions, and valued under bear, base, and bull cases. A reverse DCF backs out the assumptions implied by the current market price.

Exhibit 1. Revenue history and five-year base case
Lululemon revenue: post-COVID era history with five-year base-case projection.

Source: SEC EDGAR XBRL company facts. Market price from Yahoo Finance at run date.

18.0%
Discount rate implied by market price
vs 11.9% CAPM cost of equity
-9.2%
Implied revenue growth, 5 years
Market pricing decline
$0
Long-term debt
Net cash balance sheet
16yrs
Financial history normalized
25 XBRL line items

Summary

At $117.32, the market price implies an 18.0% discount rate against an 11.9% cost of equity from CAPM, and five years of revenue decline at -9.2% a year with margin compression. The gap between the implied and modeled discount rate is consistent with a distressed-company risk premium. Lululemon carries no long-term debt, so the model is an equity-only valuation with operating leases as the only material obligation.

Results

Historical revenue with the five-year base case, the assumptions the market price implies against historical actuals, and free cash flow to the firm under three scenarios.

Exhibit 2. Reverse DCF, implied vs historical
Reverse DCF: implied revenue CAGR, operating margin, and WACC vs historical actuals.
Exhibit 3. FCFF, bear, base, and bull
LULU 5-year FCFF projection under three scenarios.

Pipeline

Bronze

Python pulls every XBRL fact Lululemon has reported from SEC EDGAR in one API call, filters to the 25 line items the model needs, and lands them in S3. Lambda-ready for scheduled refresh.

Silver

A pandas parser normalizes XBRL tag variants across old and new accounting standards into typed columns. The same parser runs on Nike, Adidas, and Under Armour without edits. Output: 16 years of financials as Parquet.

Gold

DuckDB runs SQL window functions over the silver table: year-over-year growth, five-year rolling CAGR, margin compression versus peak, and free cash flow conversion. The SQL ports to a Databricks SQL warehouse with a one-line connection change.

Valuation

Five-year free cash flow projection discounted at the cost of equity with a Gordon growth terminal value. The reverse DCF solves for the growth, margin, and discount rate implied by the current price. Bear, base, and bull assumptions run through the same model.

Background

The valuation itself could be built in a spreadsheet. The purpose of the project is the pipeline: automated ingest from SEC filings, a reusable normalization layer, SQL analytics, and a valuation model with every assumption listed.

Personal research project on public filings. Not investment advice. Code is on GitHub.

Forecasting Multi-channel DTC health & wellness brand · 200 SKUs Shopify and Square, 365 days · June 2026

Multi-Channel Reorder Forecasting System

56 SKUs are running low, and the reorder comes to $1.18M.

A forecasting and decision tool for a direct-to-consumer health & wellness brand selling through Shopify and Square. Sales history from both channels is reconciled per SKU and run through a five-state reorder engine that produces a daily purchase queue, delivered as a Streamlit application on PostgreSQL.

Exhibit 1. Overview: KPIs, sales trend, category mix
Overview dashboard showing KPI tiles for revenue, units sold, average daily revenue, SKUs in scope, and reorder count, with a sales trend line and category breakdown.
56/ 200
SKUs flagged reorder now
This cycle
$1.18M
Recommended PO value
Grouped by vendor, MOQ applied
9.0x
Velocity spread, top vs bottom quartile
Recency weighted
365
Days of rolling history
Two sales channels

Summary

Of 200 SKUs, 56 are flagged for immediate reorder this cycle, with a recommended purchase order value of $1.18M grouped by vendor with minimum order quantities applied. Every recommendation traces to its inputs: velocity, days of supply, lead time, and outstanding purchase orders.

Results

The overview reports revenue, units, average daily revenue, SKUs in scope, and reorder count. The alerts view lists SKUs by status with days of supply, recommended quantity, and purchase order value.

Exhibit 2. Reorder alerts, filtered to action rows
Reorder Alerts page showing per-SKU status dots, days of supply, recommended quantity, and PO value, filtered to reorder_now and reorder_soon rows.
Exhibit 3. Forecast detail with calculation shown
Forecast Detail page showing per-SKU sales history, forecast projection, KPI tiles, and the math breakdown of velocity and recommended quantity.

Source: Shopify Admin API and Square API order history, synthetic for this demonstration.

Method

Sync

Python pulls daily orders from the Shopify Admin API and Square API, normalizes both to one schema of SKU, date, units, revenue, and source, and keeps 365 days of history.

Schema

Products, vendors, settings, purchase log, and forecast log in a normalized PostgreSQL schema with foreign keys and indexes. Every SKU must carry a vendor, lead time, MOQ, and unit cost.

Velocity

90-day rolling average sales per SKU with linear recency weighting and an optional per-SKU growth factor. Cold-start and dead SKUs are handled as explicit cases.

Engine

Days of supply equals on-hand divided by velocity, compared against target cover plus vendor lead time to assign one of five states. Outstanding purchase orders override to on order. Recommended quantity respects vendor MOQ.

App

Six Streamlit pages: Overview, Reorder Alerts, All Products, Forecast Detail, Purchase Log, Vendors. Filters carry across pages. Each refresh writes a forecast log row for audit.

Background

Reorder timing is usually managed in a spreadsheet, which fails when a SKU turns seasonal or a second sales channel is added. This tool replaces the spreadsheet with a model whose every output can be opened to its inputs.

Portfolio project. Figures shown are from synthetic order data. Code is on GitHub.

Next up. Healthcare: a glioma survival analysis, a brain tumor MRI classifier, and a clinical trial matcher. Finance: a Vail Resorts Monte Carlo valuation.