Hire Rajan J. - Prompt Engineer - San Francisco

At Google, I work at the intersection of LLM innovation, prompt design, and scalable GenAI systems. I specialize in crafting high-performance prompts (CoT, ReAct, few-shot) for GPT-4, Gemini, and Llama models, with a focus on improving reasoning accuracy, reliability, and production-grade consistency. I build end-to-end RAG pipelines using LangChain and LlamaIndex, integrating vector databases like FAISS, Pinecone, and Chroma to deliver fast, accurate, enterprise-scale search experiences. My work also includes designing LLM-powered applications with FastAPI/Flask and deploying them across GCP and AWS using Docker and CI/CD workflows. I lead initiatives around model safety, evaluation, and optimization—leveraging LangSmith, Ragas, and custom evaluation frameworks to reduce hallucinations, enhance semantic relevance, and maintain Google-level quality standards. I also develop modular agents, function-calling tools, and workflow automation components that support complex, multi-step reasoning tasks. From building robust NLP preprocessing pipelines to architecting hybrid search systems and large-scale knowledge ingestion frameworks, my focus is always on delivering reliable, efficient, and production-ready GenAI solutions.

Contributed to the development of high-performance LLM systems by engineering advanced prompt architectures—including few-shot, CoT, and role-based strategies—across GPT-4, Claude, and other frontier models. Strengthened evaluation pipelines through Python/TypeScript automation, synthetic data generation, and large-scale adversarial stress testing. Enhanced model security and reliability by performing jailbreak analysis, designing robust sanitization logic, and enforcing structured, compliant outputs using JSON schemas and Guardrails-based validation. Led end-to-end development of scalable RAG pipelines leveraging vector stores, embeddings, rerankers, and optimized chunking to elevate grounding and long-context reasoning. Built FastAPI-based evaluation microservices, delivered metrics frameworks (BLEU/ROUGE), and implemented comprehensive observability and versioning workflows. Established strong content-safety guardrails and collaborated with cross-functional ML, safety, and data teams to advance red-teaming practices, streamline experimentation, and scale enterprise-grade prompt engineering operations.