About Rozgar
Rozgar (rozgarjobs.in) helps freshers find jobs they can actually get. Every job is for freshers and early-career people and was posted in the last 20 days, so there are no ghost jobs. Candidates get a resume match score for every job, what is missing, interview prep notes, a referral plan, and one place to track every application.
All of this runs on Rozgar AI. Next, we are training our own model on cloud GPUs, and you will help build it.
What you will do
- Turn real resume and job data into clean, labelled training sets.
- Fine-tune open models (for example Llama, Qwen or Mistral) with LoRA or QLoRA on cloud GPUs.
- Build evaluations that show whether a new model is better than the current one.
- Work on resume-to-job matching, eligibility checks and skill extraction.
- Help serve the model in production and connect it to our Node.js backend.
- Set up a retraining loop, and check results for bias.
What you should know
- Python and PyTorch
- Hugging Face (Transformers, PEFT)
- NLP and LLM basics: embeddings, fine-tuning, prompting
- Git, and comfort with Linux or the cloud
Good to have
- A project where you fine-tuned or evaluated a model
- Kaggle, open-source work or a research paper
- Search, ranking or recommendation systems; Hindi or Hinglish NLP
- Docker, vLLM, Weights & Biases or MLflow
What you get
- Train and ship a real model used by job seekers
- Work directly with the founders, with real ownership
- Free stay and food in Bengaluru, on top of the stipend
- Cloud GPU access for your experiments
- Certificate, recommendation letter, and a chance of a full-time role
Selection process
- Resume and project screening
- Take-home task (2–3 days): build a simple resume–job match model and show how you measured it
- Technical interview on your task and one past project
- Final conversation with the founders