I have reviewed hundreds of student portfolios through TechVerse Enclave, placement clinics and hiring at SCULT INDIA. Most contain the same three projects, and reviewers skip them in seconds. The ones that get interviews are not harder; they are more honest about a real problem.

Projects that get interviews

  1. A RAG system with published evaluation numbers over a corpus you understand — recall, faithfulness, refusal rate. The numbers are the differentiator.
  2. A fine-tuned small model with a base-versus-tuned comparison on a task you can explain in one sentence.
  3. A document or vision tool with a validation layer — extraction plus arithmetic checks, or recognition plus liveness. Shows you think about failure.
  4. Anything with ten real users and a changelog driven by their feedback. A shop, a club, an NGO.
  5. A tool you built for yourself and still use. Reviewers can tell.

Projects that get skipped

  1. The chatbot with no retrieval and no evaluation.
  2. The tutorial clone with the tutorial's dataset.
  3. The dashboard with fake data and no user.
  4. The "AI-powered" app where the AI is one prompt and the rest is a template.
  5. Anything without a README that says what problem it solves in the first line.

Three things that turn any project into an interview

  • A URL. Something running, even on a free tier, beats a perfect repository nobody can try.
  • Numbers. Users, evaluation scores, latency, cost. Any honest number.
  • A trade-off, stated. "We chose X over Y because Z" is the sentence reviewers remember.

How many projects

Three good ones beat twelve. One retrieval, one fine-tuning or vision, one with real users — the shape of the roadmap. Then spend the remaining time on the READMEs, because a reviewer gives each project about ninety seconds, and the README is where that time goes.

— Pranjul Rathour, GenAI Engineer from Kanpur, India. Open to GenAI roles, hackathon judging, mentorship sessions and guest talks at any campus: pranjulrathour41@gmail.com.