Founding Data Scientist

• 8+ years of relevant experience in Data Science, Machine Learning, AI Engineering, or related technical fields.
• Hands-on experience post-training open-weight LLMs using SFT, DPO, and/or reinforcement learning techniques.
• Proven experience deploying fine-tuned language models in production environments.
• Strong proficiency in Python, PyTorch, Transformers, attention mechanisms, and modern LLM architectures.
• Experience with LoRA/PEFT, distributed training using FSDP or DeepSpeed, mixed-precision training, and multi-GPU infrastructure.
• Practical experience using vLLM or SGLang for high-throughput inference and agent rollouts.
• Experience building LLM training datasets, agent trajectories, preference datasets, and model evaluation frameworks.
• Strong understanding of model optimisation, experimentation, performance analysis, and failure diagnosis.
• Ability to work independently, solve complex machine learning problems, and take ownership in a fast-paced startup environment.

• →LLM Post-Training: Fine-tune open-weight language models such as Llama, Qwen, and Gemma using supervised fine-tuning (SFT), Direct Preference Optimization (DPO), and reinforcement learning techniques.
• →Reinforcement Learning: Design reward functions and optimise model policies using real-world task outcomes, user preferences, and agent performance signals.
• →Agent Data Pipelines: Build high-quality training datasets using agent trajectories, tool interactions, execution traces, and preference signals.
• →Model Evaluation: Develop benchmarks and evaluation frameworks to measure reasoning, tool usage, decision-making, and end-to-end task completion.
• →Training and Inference: Execute distributed model training and high-throughput inference using PyTorch, FSDP, DeepSpeed, vLLM, and SGLang.
• →Production Ownership: Deploy, optimise, monitor, and continuously improve fine-tuned models in production environments.
• →Conduct experiments with model architectures, training methodologies, and optimisation techniques to improve model quality and reliability.
• →Analyse model failures, identify performance bottlenecks, and implement improvements across the training and inference lifecycle.
• →Take end-to-end ownership of model development, from experimentation and training to deployment and ongoing optimisation.

• Experience with Proximal Policy Optimization (PPO), Group Relative Policy Optimization (GRPO), and Reinforcement Learning from Human Feedback (RLHF).
• Familiarity with reward modelling and reward-function optimisation.
• Experience developing or improving long-horizon autonomous AI agents.
• Exposure to advanced reinforcement learning techniques for improving reasoning, planning, and tool-use capabilities.

The ideal candidate is an engineer or scientist who has actually trained, fine-tuned, and shipped LLMs in real-world environments. You should enjoy solving complex machine learning challenges, experimenting with model architectures and training techniques, and building proprietary AI systems from the ground up.

You should be comfortable working directly with founders, making technical decisions, and taking complete ownership of model development in a fast-moving founding team.

• Position: Founding Data Scientist
• Experience: 8+ Years
• Location: Gurgaon
• Work Mode: Work from office, 6 days a week

Must-Have Skills: SFT, LLM

Good-to-Have Skills: PPO, GRPO, RLHF