Founding ML Infrastructure Engineer
Posted Updated
Summary
Founding engineer in Bengaluru owning the systems that train and serve Kalpa Labs' speech models: the distributed PyTorch training stack (FSDP/Megatron-style), a realtime vLLM-based inference engine, large-scale audio data pipelines, and production APIs, deploy tooling, and monitoring.
We’re hiring a Founding ML Infrastructure Engineer in Bengaluru to own the systems that train and serve our Speech models: the distributed training stack, Realtime inference engine, and the cloud services. You’ll work in a small team with tons of compute and see your ideas making it to production in days.
What you’ll do
- Own the training stack: FSDP/HSDP and Megatron-style tensor parallelism in our PyTorch framework
- Make realtime inference fast: time-to-first-audio and throughput on our vLLM-based engine, Quantization, model compilation and load testing under high concurrency.
- Build data pipelines at scale: Work on creating in-house data pipelines that process Millions of hours of audio and feed training at high throughput.
- Run the production path: API gateway and streaming WebSocket APIs, deploy tooling, logging, monitoring.
What we’re looking for
- 3+ years building ML systems or high-performance backend infrastructure.
- Hands-on depth in at least one of: distributed training (FSDP/TP/PP), LLM serving (vLLM/SGLang/TensorRT-LLM), High volume data processing, or low-latency streaming systems.
- Comfort operating in ambiguity: you can spec, build, debug, and ship.
- Speech/audio experience is a plus, not required.
