Shubham Singh

About

Backend engineer from Palantir who built and productionized an enterprise LLM-based document intelligence/search platform, evolving it into a hybrid lexical+vector retrieval system. Emphasizes reliability and cost control via strict LLM gating, robust fallback paths, and evaluation frameworks (e.g., MMLU/BLEU), plus disciplined migration practices (feature flags, dual-writes, shadow reads) to ship changes safely at scale.

Experience

Education

Key Strengths

• Designed and evolved enterprise LLM-driven document search into hybrid lexical+vector retrieval
• Production-minded LLM integration with strict gating, validation, and fallback paths
• Built end-to-end document ingestion and embedding pipelines; optimized chunking for retrieval quality
• Risk-managed backend refactors using feature flags, canary rollout, dual-writes, and shadow reads
• Deep focus on enterprise reliability via observability, latency controls, and resilience patterns
• Identified subtle retrieval-quality failure modes that caused LLM hallucinations; added confidence signals and safeguards