Coveo challenge · Mariam Al-Ansary
Search that answers.
A Pokédex built on the Coveo AI-Relevance Platform: ask a question in plain language and get a cited, generated answer over indexed species pages. Plus how it's configured, and two enterprise customers who could use the same approach.
01 · The build
Three ways in: ask, explore, dive in.



02 · Configuration
One search hub ties the whole stack together.
-
01
Source
A Web source,
pokedex-V2, crawls species pages from pokemondb.net, and only those pages. -
02
Fields
Scraping pulls type, generation, name, artwork and every form. An indexing extension handles the forms.
-
03
Pipeline
The
Pokemon Searchhub selects its own query pipeline, with query correction for misspelled names. -
04
Models
- Semantic Encoder
- Generative Answering
- Passage Retrieval
- Query Suggestions
-
05
Front end
Coveo Atomic and Headless in a Vite app, served from Cloudflare Workers.
Every search and click is logged against the hub. That analytics loop is what the models learn from.
03 · Presentations
The deep dive, and who else could use it.
Part 1 · Technical deep dive
For a technical audience
Coveo Pokédex: technical deep dive
A step-by-step setup guide: the foundations, the source and its scraping rules, the pipeline and models, and the Atomic front end, written so someone else could follow it.
Open slidesPart 2 · Enterprise customers
Ford · Vehicle integration
Vehicle Software Integration Bottleneck
Every new build powers on with hundreds of fault codes, sorted by hand. How Coveo connects traces, specs and fix history so each fault reaches the team that owns the root cause.
Open slidesExecutiv · Community
Introductions, on purpose
A community of executive women where “who can introduce me to someone at X?” reaches the member who actually has that connection, and nothing is shared unless she says yes.
Open slides