Part 1: Introduction & Project (10 min)
Q1. Tell me about your recent project in Generative AI.
They want:
- A business context (who benefits, what problem).
- Tech stack (LLM, LangChain, embeddings, vector DB, cloud, etc.).
- Your role (not “team did this” → say I implemented chunking, embeddings…).
- Outcome (accuracy, time saved, cost reduced).
Q2. What challenges did you face in implementing GenAI?
They expect examples like:
- Hallucinations → handled via RAG / prompt engineering.
- Large docs → solved via chunking + embeddings.
- Cost/latency → solved via quantization / caching.
- Data privacy → anonymization, masking, on-prem deployment.
Part 2: Technical / GenAI Concepts (10 min)
Q3. What is RAG (Retrieval Augmented Generation)? Why is it useful?
They expect:
- “RAG = combining external knowledge (retrieved via embeddings/vector DB) with LLM response.”
- Solves hallucination, up-to-date info, domain-specific Q&A.
Q4. How do you handle PDFs containing images?
They expect:
- Use OCR (Tesseract, Azure Form Recognizer, Google Vision API) to extract text.
- Then chunk → embeddings → retrieval.
Q5. Explain prompt engineering techniques you have used.
They expect examples:
- Few-shot prompting.
- Chain-of-thought prompting.
- Role/task-based prompts (“You are an HR assistant…”).
- Guardrails (“Always answer in JSON format…”).
Q6. What is an imbalanced dataset? How do you handle it?
They expect:
- Definition: one class dominates others.
- Techniques: oversampling (SMOTE), undersampling, class weights, synthetic data generation.
Q7. What is the F1 Score? Why not just accuracy?
They expect:
- Harmonic mean of precision & recall.
- Used when data is imbalanced.
Part 3: Quick Coding/SQL Check (5–7 min)
Q8. (Python) Write code to print even and odd numbers from a list.
nums = [1, 2, 3, 4, 5, 6, 7, 8, 9]
evens = [n for n in nums if n % 2 == 0]
odds = [n for n in nums if n % 2 != 0]
print("Evens:", evens)
print("Odds:", odds)
Q9. (SQL) How do you remove duplicates from a table?
DELETE FROM my_table t1
WHERE EXISTS (
SELECT 1
FROM my_table t2
WHERE t1.id > t2.id AND t1.col1 = t2.col1
);
They expect: you know ROW_NUMBER(), DISTINCT, or GROUP BY approaches.
Part 4: Behavioral / Fit (2–3 min)
Q10. Why Deloitte for GenAI?
They expect:
- “Deloitte has strong focus on AI consulting + large enterprise clients.”
- “I want exposure to diverse industries & real business problems.”
- “Opportunity to work end-to-end (strategy → build → deploy).”
Q11. How do you stay updated with GenAI?
They expect:
- “Arxiv, HuggingFace papers, OpenAI/Anthropic blogs, newsletters like TLDR AI, LinkedIn/X communities.”
- Maybe mention side projects, hackathons, or content creation (if true).
That’s a realistic 30-min flow. Deloitte wants to see:
- You understand GenAI concepts & real use cases.
- You can code basic problems.
- You can explain projects in business terms.