How should I prepare for the Razorpay LLM & RAG Fundamentals round?

First confirm with your recruiter that a llm & rag fundamentals conversation is part of your loop. Then write answers to the questions below before reading ours, mark yourself against each How it's scored section, and read the linked concepts wherever you fell short. The main Razorpay page explains what we know about the loop and what we do not.

Razorpay LLM & RAG Fundamentals interview questions

Tokens, context windows, sampling, embeddings, retrieval, fine-tuning, reasoning models and multimodality, explained at the depth an engineer respects and a CEO can follow. Questions we tagged for Razorpay come first, then more from the LLM & RAG Fundamentals track. A tag is our judgment of relevance, not a report of what Razorpay asks, so confirm the format of this round with your recruiter.

LLM & RAG Fundamentals questions tagged to Razorpay

Answers I can read

4 of 4 questions · 1 answer available

Practise this question →Read answer →

More LLM & RAG Fundamentals practice

Further questions from this round. Attempt each one before opening the answer, then check your reasoning against the rubric.

Answers I can read

15 of 15 questions · 4 answers available

Practise this question →Read answer →

Concepts for LLM & RAG Fundamentals preparation

The lessons these questions build on. Each card shows whether it is open, needs a free account or needs Premium.

Foundational

Tokens and Context WindowsWhat a token is, why input and output are priced differently, and what fits in a context window.

Foundational

How LLMs GenerateWhy outputs vary, why the first word is slow, and what each stage means for product design.

Core

Embeddings and Vector SearchMeaning as coordinates, nearest-neighbour search, hybrid with keywords, and reranking.

Core

RAG ExplainedRetrieve, augment, generate: the pipeline, the choices at each stage, and where it breaks.

Core

Fine-Tuning and LoRAWhat fine-tuning changes, the data it needs, and when it is wasted effort.

Foundational

RAG vs Fine-TuningRetrieval for knowledge that changes, fine-tuning for behaviour. Often both, in that order.

Advanced

🔒 Premium

Reasoning ModelsThinking tokens buy accuracy on multi-step problems. When to pay, and how to route to them.

Core

Classical ML DecisionsChoose models and thresholds from error costs, data quality and operating constraints.

Explore the full curriculum →

RAZORPAY LLM & RAG FUNDAMENTALS FAQ

We picked these for Razorpay's products and loop, for example "What is a token, and why should a product manager care?", "The model's JSON output breaks our parser about 2% of the time. Why, and what should we do?", "We need to extract data from 2 million receipt photos a month. Send images straight to a multimodal model, or run OCR first and send text?". A company tag reflects our judgment of relevance, not a report that Razorpay asked the question. Every question shows whether its answer is free, needs a free account or needs Premium.

Other rounds to prepare for Razorpay

Rounds where we have tagged questions for this employer.

Razorpay: AI System Design→Razorpay: Agentic Products→Razorpay: AI Product Sense→Razorpay: AI Platform & Model PM→

Continue your Razorpay AI Product Manager preparation

Premium opens every graded answer and concept lesson for six months on a single payment. Try the free answers marked above first.

Unlock all answers Back to Razorpay

AIPMInterviews is independent of Razorpay. Trademarks remain with the companies that own them.

Razorpay LLM & RAG Fundamentals interview questions

LLM & RAG Fundamentals questions tagged to Razorpay

What is a token, and why should a product manager care?

The model's JSON output breaks our parser about 2% of the time. Why, and what should we do?

We need to extract data from 2 million receipt photos a month. Send images straight to a multimodal model, or run OCR first and send text?

You're launching a new AI product for customers in the US, EU and India. How do you choose the model stack?

More LLM & RAG Fundamentals practice

Our CEO asks whether we should fine-tune a model on our documents or use RAG. Explain the difference and make a recommendation.

Why does the model give different answers to the same question, and what should a PM do about it?

How do embeddings and vector search work, and why do they sometimes return the wrong results?

New models accept enormous context windows. Should we drop RAG and just put all our documents in the prompt?

Engineering wants to fine-tune a model for our ticket-classification feature. How do you decide if it's worth it?

What are reasoning models, and how would you decide whether a feature should use one?

Why do language models hallucinate, and what can a product team actually do about it?

Walk me through what happens between a user pressing send and the model's answer appearing. What does a PM need to know?

Our RAG assistant keeps retrieving the wrong documents. How would you diagnose and fix it?

We moved to a newer model with the same price per token, but our bill rose about 30%. What happened, and how should we have caught it?

Your VP asks whether we should use an open-weight model instead of a commercial API. Lay out the trade-offs.

Our assistant costs more and responds more slowly for Hindi and Tamil users than for English users. Why, and what do we do?

Should our phone's AI assistant run models on the device, in the cloud, or both? Make the strategic case.

What is model distillation, and when would you, as a PM, ask the team to do it?

Our document assistant answers well from web pages but badly from PDFs with tables. Why, and how would you fix it?

Concepts for LLM & RAG Fundamentals preparation

Other rounds to prepare for Razorpay

Continue your Razorpay AI Product Manager preparation