Gen AI Interview

 Langchain vs LangGraph

For RAG ingestion what you have used lang chain or lang graph

For the RAG ingestion pipeline, I use LangChain, not LangGraph. 

Ingestion is a deterministic ETL workflow—load PDFs, extract text, clean it, chunk documents, enrich metadata, generate embeddings, and index them into OpenSearch or another vector database. LangChain provides mature document loaders, text splitters, embedding interfaces, and vector store integrations, making it the right choice. 

LangGraph is more appropriate for stateful, multi-step agent workflowsinvolving branching, loops, tool calls, retries, checkpoints, and human-in-the-loop interactions. In the HR chatbot architecture we've discussed, I would use LangChain for ingestion and LangGraph for the online agent orchestration(intent detection, retrieval, MCP tool invocation, approval/escalation flows, and response generation)

Torch.no_grad vs Torch.inference_mode

No Grad 

• Evaluation
• Fine Tuning

Inference 

• Inference