ML Systems Engineer

Company Description

Scanta, based in San Francisco, is a cutting-edge technology company specializing in AI and Spatial Computing solutions. Renowned for its advanced technological innovations, the company provides strategic advisory services to help clients achieve their vision. Scanta collaborates with a diverse global talent pool of product consultants and developers, delivering exceptional support and outcomes for its customers. At Scanta, we are committed to transforming ideas into impactful solutions.

Role Description

This is a full-time, on-site role located in Gurugram for a Machine Learning (ML) Systems Engineer. The ML Systems Engineer will focus on designing and implementing robust systems for machine learning applications, troubleshooting technical issues, and maintaining system operations. Responsibilities will also include providing technical support, optimizing system performance, and enhancing existing system designs to meet business and customer requirements.

Responsibilities

- Adapt optimization loop for AMD (ROCm PyTorch)

- Implement configuration generator (hyperparameter search space)

- Build a simple Bayesian optimization loop (can use Optuna or Ax)

- Implement parallel job submission (50-100 configs)

- Collect results and compute the Pareto frontier

- Support 2-3 pre-defined models (Llama-3-8B, Mistral-7B, Qwen-7B)

Key Simplifications:

- Use existing open-source optimizer (Optuna) instead of building a custom one

- Pre-define search space (don't let users customize)

- Limit to single optimization objective initially (accuracy vs speed)

- Skip compression techniques for v1 (focus on training/inference optimization)