AI Natural Language Car Search
Back to Work
Data/MLAISearch2024

AI Natural Language Car Search

Shifted from rigid filters to natural language: users can now search 'family hybrid SUV under 20k in Barcelona'.

Mauro Ferrantelli
12 November 20245 min read
On this page (6)
  1. 01Overview
  2. 02Technical Approach
  3. 03Moderation
  4. 04Cost & Scalability
  5. 05Learnings
  6. 06Next Steps

01Overview#

We set up a private 15‑day sprint to prove natural language search could cut friction and boost relevance.

02Technical Approach#

AI does not execute the search itself; it moderates the user phrase then translates it into existing structured filter parameters.

Technical flow
Fig 1

03Moderation#

We block off-topic or unsafe queries before attempting filter translation.

Moderation flow
Fig 2

04Cost & Scalability#

Token budget tuned to keep costs low while validating adoption.

Insight: Shipping a working thin slice early generated instant stakeholder buy‑in.

05Learnings#

  • Clean data matters
  • Rapid iteration with few-shot examples
  • Upfront moderation reduces cost & risk

06Next Steps#

Drive adoption by surfacing smart suggestions and lightweight education in the search box.

Impact Highlights

Automatic natural language → structured filters mapping

Fewer steps to relevant results

Reusable technical base for future conversational experiences