Machine Learning Engineer

NuPlay AI; Bengaluru; On-site; INR 2,500,000 - 5,000,000; Full time

Job summary

Nurix AI is seeking a Machine Learning Engineer to solve challenging problems in speech, LLMs, and agentic AI at scale in Bengaluru.

You will own evaluation of providers, build real-time voice pipelines, and push model quality improvements through fine-tuning and efficient inference, collaborating with MLOps and product teams.

Qualifications

• Bachelor's degree in CS/AI/ML or related field.
• 4+ years of hands-on ML engineering in speech/NLP.
• Strong expertise in ASR, TTS, NLP, LLM inference & fine-tuning.
• Proven track record deploying models to production.
• Experience with data ops for training/test data.

Responsibilities

• Evaluate and integrate ASR/LLM/TTS providers.
• Benchmark latency, accuracy, and noise robustness across languages.
• Design real-time voice pipelines with robust endpointing and variability handling.
• Improve model quality via fine-tuning and prompting.
• Collaborate with MLOps to deploy low-latency production systems.

Skills

ASR

TTS

NLP

LLM inference

Fine-tuning

Python

PyTorch

Real-time systems

Education

Bachelor's degree in CS/AI/ML or related field

Tools

PyTorch

vLLM

SGLang

Unsloth

Axolotl

Job description

As a Machine Learning Engineer at Nurix AI, you will play a pivotal role in solving some of the hardest challenges. You'll work on building robust agentic systems, designing evals and managing model deployment at scale. This role is ideal for engineers who want to push the boundaries of speech, LLMs, and agentic AI in production at scale.

The candidate will have responsibilities across the following functions:

Speech and Voice Systems:

• Own evaluation, selection, and integration of ASR/LLM/TTS/S2S model providers.
• Benchmark accuracy, latency, and noise robustness across language and code-switched settings, including noise-cancellation and audio pre-processing variants.
• Build and optimise real-time streaming voice pipelines that handle interruptions, endpointing, natural pauses, and real-world audio variability without breakdowns.

Model Quality and Inference:

• Analyse conversational quality of voice agents against human benchmarks and translate findings into prioritised model improvement roadmaps (fine-tuning, prompting, harness engineering).
• Design SFT datasets and lightweight eval frameworks for voice-agent quality at scale.
• Optimise LLM/ASR/TTS inference for in-house model deployments including memory/KV-cache planning, GPU selection and sizing, latency-throughput trade-offs for real-time voice workloads.
• Harden prompt, harness and tool-calling reliability for production agents.
• Work on self-learning agents capable of continuous improvement by observing human workflows and feedback

Engineering and Deployment:

• Collaborate with MLOps and Inference Providers to bring models into real-time, low-latency production environments.
• Work with DataOps to create, refine and build high-quality real-world datasets for in-house data needs.
• Ensure scalability, security, and reliability of deployed ML systems.

Collaboration and Research:

• Partner closely with product and engineering teams to deliver production-ready features.
• Stay on top of the latest advances in ASR, TTS, LLMs, speech systems and real-time inference frameworks, and apply them to Nurix's roadmap.

Requirements:

• Bachelor's degree in Computer Science, AI/ML, or related field.
• 4+ years of hands-on ML engineering experience, with a focus on speech/NLP systems.
• Strong expertise in ASR, TTS, NLP, LLM inference & fine-tuning, and conversational voice systems.
• Track Record of training/ finetuning models and deploying them in production.
• Worked closely with data operations teams for training/ testing data annotation and post-release sanity checks.
• Proficiency in Python Backends and ML frameworks (PyTorch, vLLM, SGLang, Unsloth, Axolotl).
• Experience in designing low-latency, real-time ML applications.
• Strong understanding of ML lifecycle, evaluation, and deployment practices.
• Contributions to open-source ML projects or research publications.
• Prior work on multilingual or Indian-context ASR/TTS systems.