Standard "Titanic" or "House Price" projects are getting ignored by recruiters now. I want to build something that shows I can work with modern AI. Should I focus on RAG (Retrieval-Augmented Generation) systems, or should I be looking at fine-tuning small models like Llama 3? What kind of project actually proves to a hiring manager that I understand the 2026 AI landscape?
3 answers
Recruiters in 2024 want to see "End-to-End" utility. Instead of a static notebook, build a RAG application that answers questions based on a specific niche dataset (like local government PDFs or specialized medical papers). This proves you can handle vector databases (like Pinecone or Weaviate), prompt engineering, and data ingestion. Fine-tuning is cool but expensive and often unnecessary for most business cases. Showing you can build a reliable, "grounded" AI system with RAG is the most in-demand skill right now. I built a "Legal Research Assistant" project in 2023 and it was the only thing my interviewers wanted to talk about.
Rebecca, for your RAG project, did you use LangChain or did you build the orchestration from scratch? I’m seeing a lot of debate about whether LangChain is too "bloated" for production use.
Don't forget the "Evaluation" part! Building a bot is easy, but proving it’s accurate with an "Eval framework" is what makes you a Data Scientist rather than just a hobbyist.
Cheryl is 100% right. Jordan, look into tools like Ragas or TruLens to add an "Evaluation" section to your project. It shows professional-level maturity.
Nathan, I started with LangChain but moved to a more manual setup for my final version. LangChain is great for learning the concepts, but for a portfolio, being able to explain how the vector search works without a wrapper is very impressive. If you can explain the math behind Cosine Similarity while showing your RAG app, you're golden.