The problem
The client is a major international food producer that sells in more than 100 countries. Its account managers built a pitch deck for each retail partner by hand: pull the partner's numbers from Salesforce, pick products, find approved images in Bynder, write the slides. Four to five hours per deck.
Discovery
I was embedded with the client as the main technical contact. I sat with the account team to learn what a good deck looks like, which data they trust, and where the time actually goes.
The answer was not the writing. It was the choosing: which products fit this partner, and which image is approved for this use.
Build or extend
Twin already had a workflow engine, built for grants. I prototyped the generator on it first. It broke on the assumptions underneath: grants are long documents in sections, decks are structured slides with assets and a choice on every slide.
So I rebuilt it standalone, as the sole developer, in Next.js and TypeScript.
How it works
The model reads the partner's account and order history in Salesforce and picks the products to pitch. The catalogue supplies each product's SKU, EAN and price. Bynder supplies an approved shot, with its usage notes. The template fills itself.
Every slide carries speaker notes that give the reason for each choice, with the record ids behind it. An account manager can check a deck in a minute instead of trusting it blind.
Delivery
Weekly releases and client demos. Along the way I onboarded the engineer who took over my previous role.