Calvin Iwata Memoji, smilingCalvin Iwata Memoji, celebrating with confetti
ProductEnterprise

IBM

A 72-hour hackathon MVP, a client chatbot, and the workshops that got both to a decision.

A MacBook showing the Salazar Tacos catering site with its IBM Watson chatbot open and offering Pricing, Dietary and Taco Ingredients, beside two iPhones running Harvest: a banana identified as spoiled, and a map of nearby compost drop-offs

The problem

I was a senior designer intern on IBM Global Sales, Client Engineering. In four months I shipped two products.

Both had the same hard part. A client, data scientists, engineers, and sales in one room, none of them sharing a vocabulary, against a date that did not move.

So the first job was never the interface. It was getting that room to agree on what we were building.

What I did

IBM Harvest was the internship hackathon: 72 hours to an MVP on global food waste. I ran the user research and designed the app, the visual language, and the brand.

In three days, most of that job is deciding what not to design.

The IBM Harvest proposal layout presenting the product concept and its intended user
The Harvest proposal. Three days is long enough to make one idea legible, and no longer.
An architecture map for IBM Harvest linking user actions to services and data
Architecture map for Harvest. Drawn with the engineers, not handed to them.

The capstone was a chatbot for a taco catering client, built with data science and engineering interns. It needed to answer customer questions and turn inquiries into orders.

I led the design workshops, prototyped, and built the front end.

Both got to a decision the same way. Discovery to see what we were looking at, framing to narrow it, solutioning to commit.

A three step process, discovery then framing then solutioning, run the same way on both products.

  1. Step 1 · Discovery
    Cluster what we sawOne shared description
  2. Step 2 · Framing
    One problem statementConstraints named
  3. Step 3 · Solutioning
    Effort against impactCommit on the record
The same three steps ran on Harvest in 72 hours and on the chatbot over the summer. The boards below are one step each.
A discovery workshop board with clustered sticky notes grouping observations into themes
Step 1, discovery. Everyone in the room ends up describing the problem the same way.
A framing canvas narrowing observations into a defined problem statement and constraints
Step 2, framing. Narrow it to something five people can finish.
A solutioning board mapping candidate solutions against effort and impact
Step 3, solutioning. Commit on the record, with the tradeoff visible.

Workshops, not documents. A workshop ends in a decision. A document ends in comments.

Outcome

Two MVPs in four months, both with mixed teams, both against a fixed date.

The process is the transferable part. It’s why a 72-hour build came out coherent instead of just finished.

More work

All work