When I joined The Tao of Tea — a specialty tea manufacturer and wholesaler in Portland — order fulfillment ran the way it does at most small manufacturers: manually. Purchase orders arrived as PDFs in a dozen formats. Staff retyped them into spreadsheets, hand-calculated what to pick, guessed at how to pack cartons, and wrote labels one at a time. Every step worked — and every step leaked time, invited errors, and depended on one person’s memory. Nobody asked me to fix it. I just couldn’t stop seeing it as one system with five broken links. So I built the systems that now run it daily — used by non-technical warehouse staff, fully offline, designed to outlast me.
The problem: Orders arrived from Amazon, iHerb, and wholesale channels in completely different formats. Understanding total demand — how much of each tea to produce — meant manually consolidating everything, a slow and error-prone ritual.
What I built: A tool that ingests purchase orders across channels and aggregates them into a single demand picture, so production planning starts from one reliable number instead of five spreadsheets.
The problem: Every retailer formats POs differently — and turning a PO into warehouse action meant manually working out what to pick, which retailer SKU maps to which internal product (the same tea exists as a tin, a pound bag, and an Amazon “Pack of 2”), and how to pack cases into cartons without exceeding weight limits.
What I built: A fully offline desktop app with rule-based PDF parsers tuned to each retailer’s layout, plus a standardized Excel intake for everyone else. It cross-references every line against an editable product master database — resolving SKUs, Amazon pack variants, and case specs — then runs a constrained packing algorithm that maximizes carton utilization under a strict weight ceiling. Output: a print-ready pick list and carton-by-carton packing map.
The design principle: nothing hardcoded. The product database is an Excel file staff can edit — new products, new pack variants, zero code changes.
The problem: Cases were labeled by hand — and a mislabeled case means the wrong product shipping to a customer, the most expensive error a wholesaler can make.
What I built: A label-printing application generating accurate, standardized case labels directly from product data.
Impact: Wrong shipments reduced to near zero.
The problem: Receiving inventory meant hand-writing labels for incoming shipments — slow, inconsistent, unscalable.
What I built: A desktop application deployed directly to the warehouse’s thermal printer, generating standardized incoming-shipment labels on demand — used by floor staff daily.
The problem: Every outgoing order needed a packing slip assembled manually from order data — repetitive, time-consuming, inconsistent.
What I built: A tool that generates complete, accurate packing slips automatically — manual iHerb order processing, eliminated.
End-to-end analysis of a specialty manufacturer's wholesale channel — turning raw order exports into bundle, re-engagement, and merchandising decisions.
Methods: basket/co-occurrence analysis, RFM retailer segmentation, cohort retention, Pareto concentration, and scenario modeling. The headline finding — a product trio appearing in 43% of orders — became a bundle strategy.
CNN-based detection of subtle visual artifacts in video frames — 82% accuracy across 6,000+ preprocessed frames.
Forecasting stock price movements by integrating financial market data with social media sentiment analysis.