Senior AI Engineer, LLM Deployment and Evaluation (India)

Fully remote Immediate joiners only, non-negotiable

NeuroDrift builds production voice AI for enterprise contact centres. Our platform handles 300+ production calls a day and we have shipped over 350 voice AI deployments. Models we pick and tune sit in a live call path with a real customer on the other end, where a 200ms regression is something people hear.

We are hiring one senior, hands-on engineer to own the model layer: choosing models, serving them ourselves, fine-tuning them, and proving they hold up under load.

This is a deep individual contributor role. You write code every day and you are measured on what runs in production, not on managing people.

What you'll do

• Benchmark open-weight and hosted models against real workloads on quality, latency, throughput and cost per call, and make the call on what ships
• Deploy and serve models yourself on GPU using vLLM, Triton, TGI or similar, containerized, including into a client's own cloud account
• Fine-tune and adapt models, including LoRA and QLoRA and supervised fine-tuning, and build the data pipelines behind it
• Build evaluation harnesses and datasets that catch regressions before a customer does
• Tune inference for production: quantization, batching, KV cache, concurrency and GPU utilisation
• Get all of it into a live call path alongside our voice and product engineers

Who this is for

• 3 to 7 years engineering, still writing code daily, solid in Python
• You have deployed and served open-weight models in production yourself, not only called a hosted API
• You have run structured evaluations, built your own eval sets, and made model decisions from data rather than vibes
• You have fine-tuned a model and shipped the result
• Comfortable on GPU infrastructure: memory limits, quantization, throughput tuning, cost
• You deploy and operate your own services on AWS with Docker and CI/CD

Nice to have

• Speech models: ASR and TTS, self-hosted or containerized
• Real-time or streaming inference, where .