When I review candidates for AI work at SCULT INDIA or coach students before interviews, the questions that matter are the ones a tutorial cannot answer for you. Here are the ones I would ask, grouped, with what a strong answer contains. Prepare these and you will also have built the right things.
Retrieval
- How did you decide your chunk size? Strong: an evaluation set and recall numbers at three sizes.
- Why hybrid retrieval? Strong: an example query where vectors failed and keywords worked.
- What does your system do when nothing relevant is retrieved? Strong: a threshold and a refusal, tuned on unanswerable questions.
- How do you know your citations are correct? Strong: markers resolved to chunks, a faithfulness check.
Fine-tuning
- When would you fine-tune instead of using retrieval? Strong: behaviour versus knowledge, with an example of each.
- What did your loss curve look like, and when did you stop? Strong: eval loss, checkpoints, the best step.
- How did you prove the tuned model was better? Strong: base versus tuned on held-out data, same rubric.
- What was wrong with your dataset the first time? Everyone has an answer; the honest one is the strong one.
Reliability and cost
- What happens when the provider is down? Strong: fallback, retries with limits, a typed error to the client.
- How do you control cost per request? Strong: context caps, caching, cheaper routing for simple cases.
- What do you log for an LLM call? Strong: request id, model, tokens, latency, retrieval scores — never raw personal data.
- How would you test this system in CI? Strong: golden prompts, schema checks, an evaluation script with thresholds.
Product judgement
- Was AI the right tool for this feature? Strong: names the part done with plain code.
- What did you cut, and why? Strong: a deliberate scope decision.
- Who used it, and what did they say? Strong: named users and a change made because of them.
- Where does user data go? Strong: a one-sentence data flow, and what is never stored.
About you
- Show me something running. The best candidates have a URL ready.
- Walk me through the hardest bug. Reveals who built what.
- What would you build differently now? Reveals reflection.
- What are you learning this month? Reveals momentum.
Notice that every strong answer is evidence from something you built. That is the whole strategy: build the projects in the roadmap, and the interview prepares itself.
GenAI engineer and AI product builder with 2+ years shipping production-grade AI systems: RAG pipelines, fine-tuned LLMs, hybrid retrieval and multi-modal apps across vision, speech and OCR, architected end to end from ingestion to deployment. Leads engineering for SCULT INDIA's 14-member team, founded the 500+ member TechVerse Enclave community and has mentored 200+ students. Three hackathon first prizes: Changethon 2025 (IIT Roorkee), Product Genesis at Vividhotsava 2025 (CSJMU Kanpur) and BYTEBATTLE (MeetKats).
Open to GenAI roles, hackathon judging, mentorship sessions and guest talks at colleges. Email pranjulrathour41@gmail.com.
