Job Description
About Frontier
Frontier is a subsidiary of Fresh Prints. Fresh Prints is a New York based, fast-growing, fully-remote, 150-person company that has most of our team in India and Philippines. A few years ago Fresh Prints started helping other fast-growing companies build their teams. We got so good at it that we decided to spin out a new company to focus exclusively on that and call it Frontier.
Here at Frontier, we help companies grow full-time, cross-functional teams abroad. We hire the smartest people, and we place them into the best companies. If you like one of the Frontier roles, and you apply, and you’re accepted, we’ll screen you with a couple of internal interviews, and will work on getting you an interview for a full-time job within the month. Think of us as your personal talent agent, and good luck with the application :)
About the client
Balto’s mission is to power a new era of knowledge work in the contact center, and we’re creating awesome technology to do just that. If you’re excited by the opportunity to join a dynamic team initiating a technological revolution in Real-Time Guidance, Balto is for you.
More than just a company, Balto is a community. A community committed to empowering each of our members. This mission is at the heart of our organization. As a member of our Technical Success team, you’ll act as the voice of the user, drive efficient implementation & product engagement from onboarding to continued adoption, and investigate solutions for complex technical issues.
ML Engineer — Signal Processing / ASR
We are looking for a hands-on Machine Learning Engineer with strong experience in signal processing, ASR, and production ML systems. This role is for someone who has built and deployed real-world audio or speech ML systems, not just experimented with models in notebooks.
You will work on real-time and post-call speech systems where accuracy, latency, reliability, and operational robustness matter.
Responsibilities:
- Design, build, and improve ASR, audio, and speech-related ML systems for production.
- Develop signal processing pipelines for noisy, compressed, telephony-style, or real-world audio.
- Train, fine-tune, evaluate, and deploy models for ASR, audio classification, diarization, redaction, or related tasks.
- Own ML workflows end-to-end: data preparation, model training, validation, inference, monitoring, and iteration.
- Optimize inference for latency, throughput, cost, and reliability.
- Debug model quality issues through data analysis, targeted evaluations, and production monitoring.
- Collaborate with product and engineering teams to turn business problems into practical ML solutions.
Requirements:
- At least 5 years of hands-on experience deploying ASR or other ML systems in production.
- Strong background in signal processing, speech recognition, audio ML, or telephony/audio pipelines.
- Experience with production ASR systems, streaming inference, VAD, noise handling, diarization, speaker/channel issues, or similar speech technologies.
- Strong Python engineering skills and experience building production services.
- Experience with frameworks such as PyTorch, TensorFlow, JAX, ONNX Runtime, or similar.
- Experience deploying models with Docker, Kubernetes, FastAPI, Triton, vLLM, TorchServe, custom inference services, or cloud ML platforms.
- Strong understanding of model evaluation, regression testing, observability, latency, memory, GPU/CPU utilization, and cost-performance tradeoffs.
- Comfort working with messy real-world data, noisy labels, domain drift, and ambiguous production issues.
Nice to Have:
- Experience with real-time ASR, call-center audio, VoIP, or telephony systems.
- Experience with Whisper, NVIDIA NeMo, Kaldi, wav2vec, HuBERT, Conformer, RNN-T, CTC, or transformer-based ASR.
- Experience with PCI/PII redaction, compliance-sensitive ML systems, or privacy-preserving workflows.
- Experience optimizing inference with ONNX, TensorRT, quantization, distillation, batching, or GPU serving.
- Experience with LLMs, RAG, embeddings, rerankers, or prompt-based systems layered on ASR transcripts.
Ideal Profile:
The ideal candidate is a practical ML systems engineer with strong intuition for audio, speech, and production behavior. They can take vague issues like “ASR quality dropped“, “latency spiked,” or “redaction missed edge cases” and turn them into clear investigations, measurable experiments, and production improvements.
