RAG projects for your resume, from 46 fresher job listings in India

What RAG jobs for freshers in India ask for

RAG jobs open to freshers in India require RAG in 19 of 46 listings, and jobs require it far more often than internships: in 16 of 25 against 3 of 21. On 29 October 2026 we read Naukri's search for RAG jobs at 0 years of experience (Naukri, RAG, 0 years), which reported 129 listings. Of the first 60, we kept the 46 in software, data and AI that were not senior titles or repeats, and read each description by hand: 25 jobs, 20 internships and 1 trainee post.

Five of the 19 that require RAG ask in their own text for more than a fresher, from one to five or more years, or say "No freshers please", though Naukri lists each from 0 years. In the 41 that are open to freshers, RAG is required in 14, the job's work in 12 and good to have in 15. Naukri notes that 27 of the 46 descriptions were "aggregated from external source". Forty-six listings are a small sample from one job board on one day, and a listing's requirement is not a hiring decision.

RAG is one skill among many in AI engineer jobs, not the whole job. When we read 44 AI engineer listings open to freshers on 19 October 2026 (Naukri, AI engineer, 0 years), 14 named RAG in their description or key-skills tags. Where RAG sits in the order to learn is in our AI engineer roadmap for freshers in India.

What the RAG work in these jobs involves

The RAG work these 46 fresher listings describe is mostly a vector database and embeddings, with an API around them. Retrieval-augmented generation was named in a 2020 paper by Patrick Lewis and 11 co-authors, whose models paired a pretrained sequence-to-sequence model with a dense vector index of Wikipedia (Lewis et al.). The listings name the same two parts in today's tools: a large language model and a vector database.

Ten of the listings name a vector database product: Chroma in 8, Pinecone in 7, Weaviate in 5, FAISS and pgvector in 4 each, Qdrant, Milvus and OpenSearch in 2 each, and Azure AI Search in 1. These are counts, not advice: no product is named in more than 8 of 46. The finer parts of RAG, such as hybrid search, re-ranking and citations, are named in one listing or none.

RAG projects for beginners

A first RAG project for a resume can be built from the parts these fresher listings name most. The two projects below are ours, not projects any listing asked for.

Project 1: questions answered from your own documents, as an API. Take a set of documents you are free to share, such as public documentation or your own notes. Split them into chunks, turn the chunks into embeddings, store them in a vector database, and for each question retrieve the closest chunks and have a large language model answer from them. Serve it as an API that someone else can call. Of the 46 listings, 26 name a vector database, 17 embeddings, 39 Python and 33 an API, FastAPI by name in 13.

Project 2: the same project, evaluated. Write an evaluation set: questions, each with the passage that should be found and the answer that should come back. Then measure whether retrieval finds the passage and whether the answer keeps to it. The Ragas paper names three things to measure: whether retrieval finds "relevant and focused context passages", whether the model uses them "in a faithful way", and the quality of the answer itself (Es et al., Ragas). Evaluating the output is asked in 18 of the 46 listings, evaluation sets by name in 3, and hallucination in 4.

Advanced RAG projects

Advanced RAG projects go one of two ways. Better retrieval goes past what most fresher listings name: hybrid search and re-ranking are each named in 1 of the 46, so it is depth, not a requirement. RAG inside an agent is closer to the listings: agents are named in 30.

Project 3: better retrieval, measured. Add keyword search beside embedding search, add a re-ranker, and run your evaluation set before and after each change. Anthropic measured steps like these on its own datasets in September 2024: the share of relevant passages missing from the top 20 retrieved chunks fell from 5.7% to 3.7% with contextual embeddings, to 2.9% adding contextual BM25 keyword search, and to 1.9% with re-ranking. Its conclusions include "Embeddings+BM25 is better than embeddings on their own" and "Reranking is better than no reranking" (Anthropic, Contextual Retrieval). That is one company's measurement on its own data; the numbers that matter in a RAG project are the ones from its own evaluation set.

Project 4: RAG as a tool of an agent. Give an agent the retrieval step as one tool among others, so that it decides when to search the documents and when to answer directly. Of the 46 listings, 30 name agents, agentic workflows, tool calling or MCP, and 24 of those put them in the job's work or its requirements. LangGraph is named in 15.

A RAG project for resume screening

A RAG project for resume screening reads the phrase the other way: a system that finds the resumes matching a job description and has a large language model explain the match. None of the 46 fresher listings asks for this project, so it is a choice of subject, not a requirement.

Build a resume-screening RAG only on public sample resumes or ones you make up, never on real people's resumes. Store each resume's fields, such as skills, years of experience and location, as metadata, so a search can filter on them before it compares meaning. Then evaluate it the way Project 2 does: write job descriptions whose matching resumes you already know, and measure how many the system finds. The evaluation results are what make a resume-screening RAG more than a demo.

Which framework for a RAG project: LangChain, LlamaIndex or LangGraph

The RAG listings for freshers name LangChain more than any other framework. Of the 46 on Naukri in October 2026, 25 name LangChain, LlamaIndex or LangGraph: LangChain in 21, LangGraph in 15 and LlamaIndex in 11. Hugging Face is named in 7, CrewAI in 5 and AutoGen in 4.

Twenty listings name a large language model by vendor or name: OpenAI or GPT in 17, Gemini in 14, Claude or Anthropic in 11 and Llama in 7. No single framework or model is named in every listing. Saying in the project's README which framework and model it uses, and why, gives the reader the decision as well as the code.

What to show with a RAG project

Most RAG listings for freshers do not require a project, and none says what a project does for an application. Of the 46, 6 require built work to show or explain, 7 count projects as experience or evidence, 12 list them as a plus, and 21 do not mention them. One requirement reads "at least one LLM-integrated fullstack application or agent you can show us (GitHub or live demo, not a tutorial follow-along)". Another, a listing that also asks for three or more years, wants candidates "who can clearly walk through projects they have built, including the architecture, technologies used, key technical decisions, challenges faced, and the outcomes achieved".

Two listings name a project that uses RAG, both as an advantage: "A project using LLM APIs, RAG, agents or machine learning (final-year project, hackathon or GitHub portfolio)", and "Exposure to Generative AI or RAG through projects, coursework or self-learning is an advantage". GitHub is named for the candidate's work in 5 listings, a demo in 2.

A RAG project made for these listings is one someone else can open and check: a repository that shows the architecture and how to run it, the evaluation set and its results, and a demo link. Fresher full-stack AI engineer listings ask to see the same kind of LLM project, as our guide full-stack AI engineer in India shows.

Do RAG jobs ask freshers to train models?

Most RAG jobs for freshers in India do not ask the candidate to train models. Of the 46 RAG listings on Naukri in October 2026, 6 put training or fine-tuning machine learning models in the job's work or its requirements, and 3 more list it only as a plus. Most of the rest ask for the work around a model someone else trained: retrieval, vector databases, prompts, agents and the API.

That is the AI engineer side of the split in our guide AI engineer vs ML engineer: the ML engineer trains models on a company's own data, and the AI engineer builds the software around a model someone else trained. A RAG project shows the second kind of work.

Where Prokopi Vanguard fits

Prokopi Vanguard's curriculum, as its program page lists it, has a module called "Embeddings, Vector DBs & RAG", described as "Retrieval-Augmented Generation end-to-end" over "~5 weeks". The program page also says: "Build a production RAG engine and an autonomous AI agent — two deployed, resume-ready capstone projects." The program, its modules and its terms are on the Prokopi Vanguard page.