Winning entry — Porsche Digital Campus Challenge
An AI-powered customer relations platform for Porsche: a multi-modal assistant that resolves owner issues in the My Porsche app using semantic retrieval over Porsche's own knowledge sources, paired with an agent console that gives human CR staff the full session context the moment a case escalates. Our team took first place in the Customer Relations track.
Porsche asked how customer relations could feel as considered as the cars. The honest problem underneath: owners wait for answers that already exist somewhere in Porsche's documentation, and when they finally reach a human, they have to start the story over.
Most teams in the room built a better chatbot. We argued that the interesting failure is not the answer — it is the handover. A luxury service experience is not one where you never meet a human; it is one where meeting a human costs you nothing you already spent.
So we designed the escalation path first and the assistant second. Everything the agent knows travels with the case: transcript, retrieved sources, the vehicle context, and how confident the system was when it gave up.
Intake
Text
Image
Voice
Agent
Intent + context
vehicle, history, warranty
Semantic retrieval
manuals · service records
Tool lookups
booking · parts · status
Resolution
Guided answer
cited, step by step
Satisfaction check
resolved or not
Human handover
full context preserved
Staff
CR console
queue · KPIs · sessions
Intake is multi-modal, retrieval is grounded in Porsche's own sources, and escalation carries everything with it.
Won the Customer Relations track of the Porsche Digital Campus Challenge.
Designed the agent workflow end to end: multi-modal intake, semantic retrieval, tool-based lookups, satisfaction check, then human handover.
Built retrieval over Porsche manuals, service records and support history so answers cite the owner's actual vehicle rather than generic documentation.
Made escalation lossless — the human agent inherits the full transcript, retrieved sources and the assistant's own confidence signals.
Designed the CR staff console: live session tracking, resolution KPIs and a queue ordered by urgency rather than arrival.
Including what each one cost. A decision without a stated trade-off is usually a decision that was never really made.
Semantic retrieval over manuals, service history and support records instead of relying on the model's own knowledge.
Why
A wrong answer about a brake system is not a UX problem. Retrieval makes answers checkable and lets the assistant cite the exact document a human agent would have opened.
The assistant hands over when its own retrieval confidence drops, before the customer has to ask for a person.
Why
Waiting for the customer to give up is the experience we were asked to fix.
