Note: The job is a remote job and is open to candidates in USA. Evolution Cloud Services (EVOCS) is an IT consulting firm that helps businesses improve performance and achieve objectives through technology solutions. The Senior Machine Learning Engineer will lead the data, model training, evaluation, deployment, and monitoring efforts for AI initiatives while making model design decisions and mentoring engineers.
Responsibilities
- Be the primary person responsible for the data piece of our AI initiatives. This could be called the unglamorous stuff that decides whether the model works, but this is your passion. You put the enthusiast in MLE
- Build and maintain training and retraining pipelines in Azure AI Foundry. We’re talking working on the model catalog, fine-tuning workflows, deployment, drift monitoring, and closing the loop when production data reveals the eval set was lying to you
- Make the real model design calls because you know best which way to go: full fine-tune vs. LoRA/QLoRA vs. DPO vs. “better prompting would save us three weeks.” Know when not to train
- Run hyperparameter work that isn’t a grid search copied from a 2021 Medium post
- Operate distributed training setups and know what breaks at scale. Pick your poison: FSDP, DeepSpeed, Megatron, accelerate, etc
- Design eval harnesses that catch what’s actually wrong, with a skeptical eye on benchmark contamination
- Ship models into production as the load-bearing piece of the product, not a feature slapped on the side
- Mentor engineers who can call an inference endpoint but have never trained one themselves
Skills
- 5+ years of ML engineering experience, with meaningful time spent fine-tuning transformer models end-to-end. We're not talking notebook demos, we mean real runs with real eval harnesses where you worked through and found the problems and fixed them
- Strong Python and PyTorch, plus fluency with the Hugging Face stack (transformers, datasets, accelerate, peft, trl)
- Azure AI Foundry experience (or strong Azure ML adjacent and willingness to get deep), plus SQL and at least one data pipeline tool (dbt, Airflow, Dagster, or Spark. We're not religious)
- Experiment tracking discipline (W&B, MLflow, or a spreadsheet you defend philosophically) and the usual engineer stuff. You should know Git, Docker, and have the ability to actually ship
- Fluent in English (written and spoken) – bilingual or near-native level
- Strong interpersonal and communication skills – this is a client-facing role that involves frequent interaction via email, calls, and meetings
- Bonus for JAX; extra bonus for having read a CUDA kernel and not flinched
- You've already built or seriously operated a distributed training setup and know how to set that up
- Run quantized models locally — GGUF, GPTQ, AWQ, MLX — and know what K-quants are and why Q4_K_M is usually the sweet spot
- Familiarity with the whisper.cpp / flash-moe universe — efficient inference on hardware that shouldn't be able to do that. (Spoiler: our projects will go here.)
- A strong take on MoE routing, speculative decoding, or why KV-cache management is more interesting than it has any right to be
- RLHF, DPO, or preference data curation experience
Company Overview