I’m a Staff Applied Scientist and Head of AI at Omneky, building production-grade generative AI systems for creative advertising. I design and ship multi-agent LLM pipelines, RLHF workflows, and scalable model infra that combine text, images, and audio to generate brand-aligned ad creatives and measurable uplift.
- Lead a small cross-functional team to build end-to-end generative advertising products (modeling, data pipelines, infra, and A/B testing).
- Work on LLM fine-tuning (LoRA / QLoRA / PEFT), RLHF (PPO / GRPO), reward modeling, and evaluation.
- Design infra for large-scale training and inference: DeepSpeed / FSDP, Accelerate, Ray, Kubernetes (EKS), autoscaling, and GPU scheduling for A100 / H100 clusters.
- Integrate multimodal pipelines (image + text + audio), vector search (Pinecone), feature stores (Redis), and production asset workflows (S3).
- Multi-agent orchestration for creative workflows (LangGraph / custom agent stacks).
- Better reward models and evaluation for brand alignment and CTR/CPA optimization.
- Memory- and cost-efficient model formats and paged-KV inference (vLLM, MXFP4-style quantization research).
Other tools I use regularly: PyTorch, Hugging Face, DeepSpeed, Accelerate, Ray, Docker, Kubernetes, AWS (S3, ECR), Pinecone, Redis, MLflow, vLLM.
- Master’s in Computer Engineering (NYU) — machine learning & systems focus.
- Bachelor’s in Computer Science (India).
- Published research and engineering work in generative models and scalable ML infra.
I create practical, example-driven content about data science, LLMs, and production ML—short tutorials and deep dives on LinkedIn and GitHub.
I’m originally from India. Outside of work I enjoy the gym, hiking, playing guitar, discovering music, and making cocktails.
- LeetCode: https://leetcode.com/sidakw/
- LinkedIn: https://www.linkedin.com/in/inderpreet-singh-walia-899328132/
- StackOverflow: https://stackoverflow.com/users/16266028/inderpreet-singh
- Kaggle: https://www.kaggle.com/inder123