← Back to projects
Eurowings · Lufthansa Group•4-month production rollout
AI-Powered Semantic Search
Enterprise search powered by embeddings, retrieval, and LLMs
Built an onsite search and FAQ discovery platform for airline e-commerce, handling natural-language search, multilingual queries, and hybrid retrieval at customer scale.
Semantic SearchRAGVector SearchGPT-4LlamaIndexDatabricks
Business Problem
Keyword-only search caused poor retrieval for natural-language questions, multilingual queries, and support intent inside booking journeys.
Architecture
- Nuxt.js website and backend search API
- Databricks query understanding for language detection, spelling correction, and intent extraction
- Hybrid BM25 + vector retrieval with confidence scoring and reranking
- LLM summarization for high-confidence answer generation
Technical Decisions
- Used hybrid retrieval because vector search alone created false positives and keyword search alone missed semantic intent.
- Kept retrieval deterministic and used GPT-4 only for final summarization to control cost and hallucinations.
- Used Databricks Vector Search to align with existing governance, security, and deployment workflows.
Production Challenges
- Reduced latency with caching, parallel retrieval, prompt optimization, and top-k tuning.
- Introduced confidence thresholds and score-gap heuristics to handle vector-search false positives.
- Added offline golden-set evaluation and canary rollback after an embedding-model update hurt recall.
Outcomes
18% reduction in customer support calls
45% increase in search satisfaction
40% reduction in average latency
Higher self-service engagement for baggage, cancellation, and check-in content