AI/ML Technology Architect - DaAI

AI/ML Technology Architect - DaAI

• Hands-on experience designing and shipping LLM-powered or agentic AI systems in production, not limited to notebooks, PoCs, or isolated demos. • Demonstrated experience with multi-agent orchestration in production, using frameworks such as LangGraph, AutoGen, CrewAI, LangChain, or equivalent technologies. • Proven experience building SQL or structured data agents that can connect to live databases, inspect schemas, infer semantic meaning, and generate relationship-level understanding. • Strong working knowledge of RAG, vector databases, embedding models, chunking strategies, hybrid retrieval, metadata filtering, prompt engineering, and LLM evaluation. Deep knowledge of Pinecone, Milvus, or Qdrant, specifically around hybrid search (sparse + dense), reranking models (Cohere/BGE), and dynamic chunking strategies. • Experience deploying open-source models (Llama, Gemma) via vLLM or Ollama to optimize throughput and cost. Good to Have • Experience with knowledge graphs, ontologies, semantic data models, or enterprise metadata models. • Open-source contributions in the AI/ML, data engineering, or agentic AI ecosystem. • Experience with MLOps, LLMOps, model monitoring, observability, and production AI governance. • Exposure to custom model training, fine-tuning, or domain adaptation, though the platform will primarily build on API-based and open-source LLMs. • Architect production-grade multi-agent AI systems using LangGraph, AutoGen, CrewAI, or equivalent orchestration frameworks. • Design stateful agent workflows • Define agent capabilities for data discovery, profiling, scoring, enrichment, intelligence extraction, and contextual reasoning across enterprise data estate. • Build and guide the design of structured data agents that can introspect live databases, infer schema meaning and generate ER-level understanding. • Design document intelligence pipelines for large-scale extraction from unstructured data like PDFs, Word documents, emails, call transcripts, and semi-structured enterprise content using tools such as Azure Document Intelligence, AWS Textract, LlamaParse, or equivalent technologies. • Architect vector database and retrieval pipelines, including chunking strategies, embedding model selection, metadata design, hybrid search, retrieval tuning, and domain-specific RAG patterns. • Define agent evaluation methodology covering accuracy, precision, recall, recall@k, regression testing, drift detection, hallucination checks, and robustness testing for non-deterministic AI outputs. • Establish AI safety and trust patterns, including semantic guardrails, jailbreak protection, prompt injection, data exfiltration prevention, toxic output mitigation, policy-based response control, and secure tool-use design. • Architect agent communication and message queuing patterns using RabbitMQ, Apache Kafka, or equivalent messaging platforms for scalable and resilient agent-to-agent/task communication.