Mindfire Solutions Software Engineer AI/ML interview questions

DSA questions - compress string (aaabbc --> a3b2c1) - char frequency in string - perform left rotation on a list by d, without using additional list Gen AI questions - Talk about recent gen ai project - Detailed question about RAG architecture ( pre processing, chunking, embedding, vector db) - How exactly storage and retrieval happens in vector db - Type of vector db and when to use which one - in memory vs persistent vector db - semantic search in detail, formula of cosin similarly - How to handle caching, if user uploads same file in a chatbot, how to handle - HNSW, IVF, what is the use and their working - HNSW available in which db - how you pass chat history to llm - how is chat history passed to llm in langchain - short term and long term memory, how to implement in langchain - how to build custom tools, in langchain / dspy/ crewai - tool decorator - how tool decorator uses docstring of tool function - Under the hood how tool decorator works - how to pass info like session ID to tool, and also not expose it to llm - how to pass tools to an agent - if a value is not passed by llm how to get that in the tool, will it fail? - LangGraph basics - Pydantic and its uses - what llms have you worked with (list all models / versions, eg: gpt 4 --> gpt 4.1 , 4o, 4 mini etc, gemini--> 1.5 flash 1.5 pro etc) - open source models - hugging face vs ollama - How to handle a CSV file in RAG - how to handle the table data (unstructured.io) - have you tried mistral for extracting table? - pypdf, pdfplumber and what other libraries have you tried for table extraction, which works best? - how to make image embeddings, what model is used - multimodal llms (take image/ audio/video/ file as input, and gives multimodal output) - Talk about any other project of yours - Few questions related to project - Have you also worked on ML models apart from gen ai? - How do you evaluate RAG? - What is precision and recall in RAGAS, also in ML, formula and definition - Any questions for me..