Mlops engineer

Acura Solutions; Mumbai; On-site; INR 600,000 - 1,200,000; Full time

Job summary

Acura Solutions seeks an experienced MLOps Engineer to build and maintain high-throughput inference endpoints and configure orchestration layers for production AI models.

You will optimize models for edge devices, quantize with advanced techniques, create local data pipelines for on-premise fine-tuning, and implement CI/CD processes with MLflow.

Qualifications

• 1+ years of MLOps experience deploying, scaling, and optimizing models in production.
• Experience with Triton Inference Server, vLLM, TensorRT-LLM, and ONNX Runtime.
• GPU memory management, CUDA kernel optimization, LoRA weight swapping, and cold-start mitigation.
• Python, Docker, shell scripting, MLflow/Weights & Biases for model lineage and versioning.

Responsibilities

• Production model serving endpoints with orchestration and model backends.
• Optimize and quantize models for edge hardware using advanced techniques.
• Build secure local data pipelines to capture failed inferences for on-premise fine-tuning.
• CI/CD for ML: package, test, deploy model artifacts into production-ready containers with MLflow.

Skills

MLOps experience

Inference runtimes

GPU optimization

Python

Tools

Docker

ONNX Runtime

MLflow

Weights & Biases

Job description

Role & responsibilities

Core Responsibilities

• Production Model Serving: Build, configure, and maintain high-throughput inference
  endpoints using advanced orchestration layers and model backends.
• Model Optimization & Quantization: Apply advanced compression techniques (such
  as AWQ, GPTQ, and FP8 native quantization) to large open-weight models (like
  Qwen2.5-VL and Llama-3-8B) to make them run smoothly on localized Edge
  hardware or low-power nodes.
• Data Flywheel Engineering: Build secure, privacy-compliant local pipeline loops that
  capture failed inferences or edge cases on-premise, allowing clients to aggregate
  data locally for future fine-tuning blocks.
• CI/CD for Machine Learning: Automate the packaging, testing, and deployment of
  fine-tuned model artifacts into production-ready containers, integrating with offline
  registries like MLflow.

Required Technical Skillset

• Experience: 1+ years of pure MLOps experience deploying, scaling, and optimizing
  large language, vision, or speech models in production.
• Inference Runtimes: Deep, hands-on production experience with Triton Inference
  Server, vLLM, TensorRT-LLM, and ONNX Runtime.
• GPU & Hardware Optimization: Proven ability to manage GPU memory allocations,
  optimize CUDA kernels, handle multi-LoRA weight swapping on the fly, and mitigate
  cold-start latencies.
• Tooling & Pipelines: Mastery of Python, Docker, shell scripting, and MLflow/Weights
  & Biases for model lineage and artifact versioning.