Description

You will own the development and optimization of multimodal AI systems, focusing on vision-language models and scalable inference pipelines.

This role is on-site.

Responsibilities

• Evaluate pretrained Vision-Language Models on domain-specific datasets and define appropriate evaluation metrics.
• Implement parameter-efficient fine-tuning techniques (LoRA/QLoRA) under limited data and compute constraints.
• Architect scalable inference pipelines for multimodal models optimized for low latency, high throughput, and cost efficiency.
• Build serving layers with API/service capabilities using reproducible environments for model deployment.
• Analyze dataset characteristics and model behavior to identify biases and failure modes.

Required Skills

• 5+ years of industry experience in ML/AI systems.
• Strong proficiency in Python and PyTorch.
• Hands-on experience with Vision-Language Models (e.g., LLaVA, BLIP, Flamingo-style architectures).
• Knowledge of parameter-efficient fine-tuning methods (LoRA, QLoRA).
• Experience with model optimization techniques: quantization, batching, and embedding reuse.
• Ability to design controlled experiments comparing baseline vs. improved models across accuracy, latency, and cost.
• Experience with Docker and containerized deployments.

Preferred Skills

• Exposure to large-scale or real-world datasets.
• Understanding of ML system design for inference scaling and optimization.

Key Skills

Education

Any Graduate