The LLM Thought a Dollar Was Still ₦450: Building a Car Pricing Engine for a Market With No Data

Short summary

A case study of building AutoValue, an AI car valuation engine for Nigeria, where LLMs systematically mispriced vehicles because training data reflected pre-devaluation naira rates. The fix was architectural: fetch live market prices via web search first, inject them as hard constraints, and let the LLM only apply relative adjustments (mileage, condition) rather than generate absolute prices.

• •LLMs recall stale prices from training data with confident fluency, especially in economies with currency devaluation
• •Prompt engineering cannot fix systematic training-data bias in absolute numeric outputs
• •Architectural fix: use live web search as a price anchor and restrict the LLM to relative adjustments only

Generated with AI, which can make mistakes.