AS
Job Description
About the Role'
'GenAI Architect'
'Responsibilities'
'GenAI Architect'
'Responsibilities'
- Architect and design scalable, production-grade RAG pipelines, including:
- Retriever selection and configuration.
- Semantic search optimization.
- Extractor logic and integration.
- Comparative evaluation of vector databases (e.g., FAISS, Weaviate, Pinecone) vs. traditional search solutions.
- Define and implement LLM-based agentic workflows, including:
- Tool integration.
- Memory handling (episodic, long-term, vector-based).
- Dynamic planning and decision making within multi-agent systems.
- Lead the design of LLM-based enterprise applications from data ingestion and fine-tuning to prompt engineering and output evaluation.
- Collaborate closely with product managers, ML engineers, and developers to translate business needs into robust AI-powered systems.
- Conduct performance benchmarking, cost optimization, and system design trade-offs for GenAI solutions.
- Stay updated with the latest research and trends in LLMs, RAG, and agentic reasoning.
- Strong knowledge and hands-on experience with Retrieval-Augmented Generation (RAG) including:
- End-to-end design of retriever-reader pipelines.
- Knowledge of embedding models (OpenAI, Cohere, HuggingFace, etc.).
- Trade-offs between vector stores and semantic indexing strategies.
- Solid experience with LLM agent frameworks (LangChain, Semantic Kernel, Haystack, etc.).
- Deep understanding of Tools, Memory, and Planning in agent-based systems.
- Strong grasp of vector databases and their internal architectures (e.g., ANN algorithms, HNSW, IVF).
- Ability to justify vector DB usage over traditional RDBMS with technical and performance reasoning.
- Proven experience designing complex AI/ML or GenAI architectures at scale.
- Proficiency with Python and key AI libraries: LangChain, Transformers, LlamaIndex, etc.
- Experience working with commercial and open-source LLMs (GPT-4, Mistral, LLaMA, Claude, etc.).
- Experience with fine-tuning LLMs or working with proprietary/custom models.
- Exposure to prompt engineering for diverse tasks (summarization, QA, classification, etc.).
- Understanding of MLOps and deployment strategies for LLM pipelines.
- This is not a developer role. Candidates with only application-layer experience or limited understanding of LLM internals, vector DB architecture, or agentic workflows will not be a fit. We expect this architect to drive architectural decisions, not just write code.