Requirements

- 3+ years of qualified experience in Data Science, Machine Learning, or Applied AI.
- Currently working at a product-based company.
- Previous experience being part of a founding or early-stage engineering or data science team, preferably having taken a product from 0 to 1.
- Hands-on experience post-training open-weight LLMs such as Llama, Qwen, Gemma, or similar models.
- Strong experience with SFT, DPO, and other reinforcement learning techniques, with evidence of taking trained models to production.
- Experience building datasets for AI agents, including collecting, filtering, labelling, and deduplicating trajectory and preference data.
- Experience designing evaluation datasets and frameworks to measure model performance.
- Hands-on experience with PyTorch, distributed training using FSDP or DeepSpeed, mixed precision, and LoRA/PEFT.
- Experience with high-throughput inference and rollouts using vLLM or SGLang.
- Deep understanding of LLMs and transformer architecture, including attention, tokenisation, pre-training, and instruction tuning.
- High ownership, high agency, and the ability to work independently in an early-stage startup environment.
- Open to working in-office in Gurgaon six days a week.
- Genuinely interested in joining a startup and building products from zero to one.

Responsibilities





- Build and improve models that learn from real-world consumer interactions and task outcomes.
- Own the post-training pipeline for open-weight LLMs using SFT, DPO, and reinforcement learning techniques.
- Design training systems that use agent trajectories, user preferences, and task outcomes as training signals.
- Build and maintain datasets by collecting, filtering, labelling, and deduplicating trajectory and preference data.
- Define what constitutes a successful task outcome and translate it into measurable training rewards.
- Build evaluation frameworks to measure model performance and decide when internally trained models can replace external frontier models.
- Run high-throughput rollouts and inference using vLLM or SGLang.
- Own distributed training workflows using PyTorch, FSDP, DeepSpeed, and multi-GPU infrastructure.
- Work closely with the founders and engineering team to improve model intelligence, reliability, and decision-making.
- Establish the model training and evaluation foundation as part of Valo's founding team.

Job Details

Gurgaon (In-office, 6 days a week)

Interview Process

- Technical Screening
- ML/LLM Engineering Round
- Model Training and Research Round
- Founders' Round
- Culture Fit Round

📌 Data Scientist (Gurugram)
🏢 ClanX
📍 Gurugram