I find the story inside supply-chain data — then build the tool that fixes it.

Every segment of every pack sequence, spelled out — "S = simplified pack," not PAC+1+S. Nothing hidden: a missing segment shows up as a red row.

Auto-fixes pack counts, splits CPS loops, recounts trailers. The despatched quantity is never touched, invented serials are flagged, and a fix that makes things worse rolls itself back.

Drop an ASN in and it's validated within seconds, logged to CSV, and sorted into _passed / _failed — no babysitting.

One click builds an executive summary, error frequencies, and corrective actions — as emailable text and polished HTML, split per supplier when a batch spans several.

A Python and OpenCV pipeline reads packaging barcodes in batches and cross-checks them against the packaging instruction and the shipping notice, while a vision model I trained myself classifies box types on the receiving line. I taught it to say "unclassified" rather than guess, which is the part that makes it worth trusting. Audit effort moves off sampling and onto the exceptions that actually matter.

I trained the camera's classification model myself — showing it every box family, tuning confidence thresholds, and teaching it to say "unclassified" instead of guessing.

Five spreadsheets, five versions of the truth. I replaced manual compliance tracking with Power BI dashboards for supplier compliance and part-packaging data, so quality, receiving and materials now argue about what to do rather than about whose number is right.

My startup's smart-shelf device read real-time inventory from load cells under the products, replacing field-rep visits for clients like ITC across remote retail locations. Three years of hardware, enterprise sales and everything in between — and where I first learned that inventory data is only worth what the physical shelf agrees with.

A bottle for hikers that heats or cools its contents with a Peltier element and an onboard battery. On a venture co-op with Northeastern's Sherman Center, I designed the concepts in SolidWorks, validated them with thermal and CFD analysis, and ran five sprints of prototyping to a 10 degrees Celsius drop on the first proof of concept.

A smart spice-storage system: modular hex-grid containers that sense their own contents with load cells and RFID, track freshness, and reorder when supply runs low. From hand sketches and brand identity through CAD to a working prototype.