David Wisdom

About

Built and deployed a production LLM system that scraped Google Maps menu photos, extracted structured prices via OpenAI, and cross-validated them against website-scraped data to automate data-quality verification at scale (replacing costly manual contractor checks). Demonstrates strong reliability instincts—precision-first prompting, output gating with image-quality metadata, and fuzzy matching/RAG techniques—plus solid orchestration (Dagster/Airflow) and observability (Sentry, Prometheus/Grafana).

Experience

Education

Key Strengths

• Built and deployed an LLM-powered menu price verification system at scale (replaced manual contractor verification)
• High-precision extraction strategy to mitigate LLM hallucinations (traded recall for precision to reduce false alerts)
• Robust prompt engineering with image-quality/disqualifier metadata to gate downstream decisions
• Strong fuzzy matching and retrieval techniques (BM25, Levenshtein, semantic similarity/RAG) to align noisy extractions to canonical items
• Modular prompt development mindset (prompt snippets/templates) enabled by LangChain/LlamaIndex
• Reliability-focused AI workflow design (observability, explicit assumption testing, graceful failure handling)
• Data/workflow orchestration experience (Dagster, Airflow)
• Effective collaboration with non-technical stakeholders to deliver usable AI-driven metrics (worked directly with CEO/sales)