Inference Systems Engineer

LambdaQ labs Pvt. Ltd.; Maharashtra; On-site; INR 1,200,000 - 1,800,000; Full time

Job summary

LambdaQ labs Pvt. Ltd. in Maharashtra is seeking a seasoned ML infra engineer to build and optimize the serving stack for fast, reliable, and affordable Vikasit Inference at India scale.

You’ll own batching, caching, routing, autoscaling, and cost management across GPU-heavy model fleets, while ensuring the OpenAI-compatible API remains rock-solid for streaming and tool calls.

Qualifications

• Deep systems engineering experience (GPU, CUDA-adjacent, or high-performance serving).
• Experience with an LLM serving framework in production.
• Comfort owning latency, throughput, and cost SLOs.

Responsibilities

• Optimize throughput and latency across model fleet (vLLM / SGLang / TensorRT-LLM).
• Build batching, KV-cache, quantization, and routing for MoE models.
• Own reliability, autoscaling, and cost of the inference platform.
• Keep the OpenAI-compatible API rock-solid for streaming and tool calls.

Skills

Deep systems engineering

LLM serving framework

Latency/throughput cost SLOs

Job description

About The Role

You’ll build and optimize the serving stack that makes Vikasit Inference fast, reliable, and affordable at India scale — kernels, batching, caching, and the OpenAI-compatible API surface developers love.

What you'll do

• Optimize throughput and latency across our model fleet (vLLM / SGLang / TensorRT-LLM)
• Build batching, KV-cache, quantization, and routing for MoE models
• Own reliability, autoscaling, and cost of the inference platform
• Keep the OpenAI-compatible API rock-solid for streaming and tool calls

What we're looking for

• Deep systems engineering (GPU, CUDA-adjacent, or high-performance serving)
• Experience with an LLM serving framework in production
• Comfort owning latency, throughput, and cost SLOs

Nice to have

• CUDA / Triton kernels
• Quantization (GGUF, AWQ, FP8)
• Multi-region infra

Sound like you?

We hire for skill over credentials. Tell us why you're a fit — links and projects welcome.