ML Engineer – Real-World Speech & ASR Systems

Kiefer; United States; Remote; USD 120,000 - 180,000; Full time

Job summary

KIEFER is seeking an ML Engineer to advance automatic speech recognition for Sophea Meet. You will work with real meeting audio, own the data-to-production loop, and iteratively improve models with product-aligned quality goals.

The role emphasizes English-language ASR initially, with Greek and other languages prioritized later. You will design experiments, assess model performance using WER/CER, and collaborate with Engineering and Product teams to deliver robust, low-latency speech solutions

Qualifications

• 3+ years of ML or research engineering experience with clear ownership and measurable results.
• Strong hands-on ASR or speech ML experience, including a model you shipped or significantly improved.
• Experience training, fine-tuning, or adapting deep learning models using real-world audio data.

Responsibilities

• Train, fine-tune, adapt, and evaluate speech recognition models based on product needs.
• Build benchmarks and regression tests for real meeting audio.
• Prepare speech datasets and investigate difficult cases like noise, overlapping speakers, long sessions, and terminology.
• Use error analysis to decide model, data, or pipeline improvements for transcription quality.
• Improve inference in accuracy, latency, throughput, reliability, and cost.
• Deploy and monitor model versions with Engineering and Product to set quality goals.

Skills

Python

ASR

PyTorch

JAX

TensorFlow

Tools

PyTorch

TensorFlow

JAX

Job description

KIEFER is seeking an ML Engineer to advance automatic speech recognition for Sophea Meet. You will work with real meeting audio, own the data-to-production loop, and iteratively improve models with product-aligned quality goals.

The role emphasizes English-language ASR initially, with Greek and other languages prioritized later. You will design experiments, assess model performance using WER/CER, and collaborate with Engineering and Product teams to deliver robust, low-latency speech solutions