AI Engineer

Join a small, high-impact team building multimodal AI systems that ship to real customers. You will own the end-to-end lifecycle—from training and evaluation to production serving—ensuring optimal latency and quality across our four production surfaces: Falcon (LLM) and Finesse (TTS/Voice Cloning).

Key Responsibilities:

• Fine-tune, evaluate open-weight models (LoRA/QLoRA, SFT), and deploy them to production without compromising quality.
• Optimize inference performance using quantization (FP8/NVFP4/INT4), speculative decoding, KV-cache strategies, and advanced batching on vLLM or SGLang.
• Build robust evaluation harnesses with benchmark suites and regression gates for continuous model improvement.
• Take ownership of latency budgets by profiling pipelines and eliminating bottlenecks down to the millisecond level.
• Bridge research and production: read papers, replicate results, validate business value, and ship solutions.
• Maintain serving infrastructure including containers, autoscaling, GPU scheduling, and observability tools.

We Are Looking For:

Required Experience:
• 2+ years building ML systems in production environments.
• Strong proficiency in Python and PyTorch; ability to read/modify model implementations rather than just using high-level APIs.
• Hands-on expertise with modern inference stacks (vLLM, SGLang, TensorRTLLM) and deep understanding of optimization techniques.
• Proven experience in fine-tuning models, proper evaluation methodologies, and interpreting metrics accurately.
• Comfortable reasoning at the GPU level regarding memory layout, precision trade-offs, and hardware bottlenecks.
• Ability to rapidly read and implement recent research papers.

Preferred Qualifications:
• Experience with speech ASR/TTS, diffusion models, image/video generation, or real-time streaming systems (WebRTC).
• Advanced quantization knowledge beyond basic scripts.
• Background in Indic language NLP and multilingual tokenizers.
• Cloud GPU deployment experience on GCP/AWS/RunPod.

Our Technology Stack:

PyTorch, vLLM, SGLang, TensorRT, ONNX, Unsloth, LLaMAFactory, TRL, Milvus, BGEM3, Redis, NATS, Docker, GCP, RunPod, WebRTC, ComfyUI.