Proficiency in Python; strong working knowledge of relevant libraries & frameworks: LangChain, LangGraph, HuggingFace Transformers, PyTorch, scikit-learn, Pandas, and NumPy.
Deep experience with LLM APIs (OpenAI, Anthropic, Google, Mistral) and open-source model deployment via Ollama, vLLM, or TGI.
Solid command of vector databases, semantic search, and knowledge graph technologies for enterprise RAG architectures.
Proficiency with MLOps tooling: MLflow, Weights & Biases, Kubeflow, or similar; experience with LLMOps tools such as LangSmith.
Strong SQL and experience with modern data platforms such as Databricks for AI-ready data preparation.
Understanding of software engineering best practices: version control (Git), containerization (Docker/Kubernetes), API design (REST/gRPC), and CI/CD pipelines.
Knowledge of how to benchmark GenAI models beyond simple accuracy (e.g., toxicity, bias, and reasoning depth).
Exposure to multi-modal AI systems incorporating vision, audio, or structured document understanding (PDFs, tables, charts).
Good understanding of the GENAI standards (MCP, A2A, A2UI etc.)
Platform Prototyping: Design and implement core ML components, such as feature stores, model registries, and automated evaluation pipelines.
Standardization: Establish best practices for the ML lifecycle, from data labeling and experimentation to CI/CD for ML (MLOps).
Scalability: Optimize model inference and training workflows to handle high-throughput, low-latency requirements.
Internal Consulting: Act as a subject matter expert for product-facing data science teams, helping them leverage platform tools to solve complex business problems.
Tooling & Automation: Build internal libraries and SDKs that simplify the transition from a local research environment to a distributed production environment.
RAG Infrastructure: Design and optimize high-performance retrieval systems using vector databases (e.g., Pinecone, Weaviate) and advanced semantic search techniques.
LLM Evaluation Frameworks: Build automated "vibe-check" replacements. Develop rigorous evaluation pipelines using LLM-as-a-judge, G-Eval, or custom scoring rubrics to measure hallucination, faithfulness, and relevancy.
Agentic Orchestration: Develop and standardize the use of agentic frameworks (e.g., LangGraph, CrewAI) to allow product teams to build complex, multi-step AI workflows.
Model Lifecycle Management: Manage the transition between model providers (OpenAI, Anthropic, Google) and open-source alternatives (Llama 3+, Mistral) through unified abstraction layers.
Cost & Latency Optimization: Implement caching strategies (e.g., GPTCache), prompt compression, and token-usage monitoring to ensure the platform remains economically viable.
Guardrails & Safety: Integrate real-time content filtering and PII masking to ensure all LLM outputs comply with corporate security and ethical standards.
