Job Summary

Job Title: ML / Fine-Tuning Engineer

Location: Hyderabad OR Pune

Notice Period: 0-30 Days

Mode of Work: Hybrid

Experience: 5+ Years

We are looking for ML / Fine-Tuning Engineer who can deliver (under supervision of ProServe Tech Lead) the end-to-end fine-tuning of open-source LLMs for a narrow, high-volume production task on AWS SFT and alignment experiments (GRPO, DPO), debugging training on multi-GPU clusters, and iterating to strict accuracy targets. Models from 8B to 70B parameters.

What We Expect

- Fine-tune open-source LLMs (Qwen, Llama) from experiment to production-ready checkpoint
- Run SFT and RL alignment (GRPO, DPO) to improve output accuracy
- Execute training on AWS GPU instances (p4d, p5, g5) using distributed training
- Diagnose/fix training issues: loss imbalances, OOM errors, gradient instabilities
- Collaborate with evaluation and data engineering to iterate on quality gaps
- Make data-driven model scaling decisions (8B 14B 70B) based on offline metrics

Requirements

- Experience: 5+ years ML engineering,



with 2+ years in LLM fine-tuning
- LLM Models: Hands-on with open-source LLMs Qwen and Llama required
- Training Methods: SFT, LoRA/QLoRA, GRPO, DPO/RLHF
- Frameworks: NVIDIA NeMo/NeMoRL, VeRL, HuggingFace TRL must have used at least two
- Distributed Training: DeepSpeed ZeRO, FSDP2, multi-node GPU orchestration
- AWS Infrastructure: p4d/p5/g5 GPU instances, SageMaker Training Jobs
- Languages: Python, PyTorch; CUDA debugging a plus

Preferred (Not Required)

Fine-tuning for tool-calling/agent tasks; multi-node GRPO/RLHF with NeMoRL or VeRL; tokenizer internals and chat-template rendering for tool-use formats. Disclaimer: This job posting has been aggregated from external source. Role details, content, and availability are subject to change. Applicants are advised to confirm the latest information directly on the company website before applying.

📌 ML / Fine-Tuning Engineer (Hyderabad)
🏢 Data Economy
📍 Hyderabad