Job Overview:
We are looking for a highly experienced AI Engineer with deep expertise in Prompt Engineering, Large Language Models (LLMs), and RAG architectures. The ideal candidate will have hands-on experience designing and deploying prompt-driven solutions in production environments and a strong command over LLMs, Python programming, and NLP evaluation techniques. This role is key to enabling intelligent document processing, secure generative AI applications, and enterprise-scale LLM integration.
Key Responsibilities:
- Design and deploy prompt-based solutions for LLM-driven applications in production environments.
- Develop, test, and optimize prompt structures to align with real-world business use cases and document requirements.
- Work with LLM APIs (e.g., OpenAI, Anthropic) and frameworks to implement context-aware and secure prompt logic.
- Leverage Python for prompt automation, data preprocessing (using Pandas or similar), and API integration.
- Implement or contribute to fine-tuning pipelines and retrieval-augmented generation (RAG) solutions.
- Evaluate prompt performance using metrics such as BLEU, ROUGE, semantic similarity, and human-in-the-loop validations.
- Address LLM-related risks including prompt injection, output filtering, and response consistency.
- Translate high-level document processing requirements into structured and logical prompt flows.
- Collaborate with cross-functional teams to align AI capabilities with enterprise needs.
Must-Have Skills:
- 10+ years overall IT experience.
- At least 2 successful LLM-based projects in production, focused on prompt engineering.
- Strong knowledge of prompt structures, context management, and token optimization strategies.
- Proficiency in Python, with experience in:
- LLM APIs and prompt engineering libraries
- Data handling with Pandas
- Hands-on experience with at least one RAG or fine-tuning pipeline.
- Familiarity with prompt evaluation techniques and metrics (BLEU, ROUGE, semantic similarity).
- Deep understanding of LLM security concerns, including prompt injection, output formatting, and limitations.
