Speech / Applied ML Engineer

About The Role

This position is a high-responsibility applied machine learning role with a focus on speech technology, designed to operate under stringent real-world production constraints. The successful candidate will enhance speech performance across Southeast Asian languages, accents, the complexities of code-switching, and noisy audio inputs while meeting strict requirements related to latency, cost, and system reliability. Entrusted with impactful production changes, the role demands maturity, initiative, and speed in execution.

Core Expectations

• Full ownership of model and pipeline enhancements that significantly improve crucial speech performance metrics.
• Transitioning experiments into fully deployed solutions independently without close supervision.
• Detection and analysis of failure cases and edge scenarios within real speech data.
• Release of models, features, or parameter tuning that demonstrably boost accuracy, robustness, or latency.
• An understanding that extends beyond standard metrics like BLEU or WER, focusing on customer and business implications.

The role requires comfort with evolving requirements, multi-lingual and imperfect datasets, balancing speed with quality and safety, and making informed decisions despite incomplete labeling data.

Responsibilities

• Experimentation and fine-tuning of speech automatic speech recognition (ASR) models tailored for Southeast Asian languages and accents.
• Design and execution of experiments under real-world production limits related to latency, computational costs, and memory usage.
• Optimization of inference processes and effective utilization of GPU resources.
• Developing approaches for handling multilingual and code-switching environments.
• Collaboration with engineering teams to integrate machine learning models into production workflows.
• Creation of evaluation frameworks and curated datasets for continuous performance monitoring.
• Comprehensive documentation of methodologies, experimental results, and design tradeoffs.

Candidate Expectations

• Founding Mindset: Focus on actual shipped improvements rather than just academic metrics; proactive evaluation of production behavior; taking ownership of speech quality; balancing research intensity with delivery speed; independent identification of model issues.
• Maturity: Clear communication on knowns and unknowns; openness about failed experiments with lessons learned; receptive to feedback from multidisciplinary teams; maintaining composure under pressure; thorough follow-up on failure investigations.
• Initiative: Suggesting new hypotheses, architectures, and data strategies; thorough root-cause analysis beyond simple hyperparameter tuning; enhancement of evaluation procedures; refinement of data curation and annotation; ongoing optimization between model performance and cost.
• Machine Learning and Speech Expertise: Proficiency in Python and PyTorch; fundamental understanding of speech and ASR technologies; hands-on experience with model training, fine-tuning, and evaluation; familiarity with GPU inference and performance tuning; practical engineering approach emphasizing application over theory.

Preferred Qualifications

• Experience working with multilingual or limited-resource speech datasets.
• Knowledge of on-device or low-latency inference deployment.
• Past involvement in shipping machine learning models into production environments.

Success Indicators

• Accountability for specific speech-related use cases or languages.
• Delivery of measurable improvements in accuracy, robustness, or latency in production.
• Identification and documentation of key failure modes along with mitigation techniques.
• Contributions to model evaluation and monitoring systems.

Benefits

• Hands-on machine learning experience applied in production scenarios with real constraints.
• Close collaboration opportunities with founders and senior engineering staff.
• Building a professional portfolio comprising experiments and deployed model improvements.
• Career growth pathway towards specialized applied ML or speech engineering roles.

Who Should Avoid Applying

• Those interested solely in working with simplified datasets and offline benchmarks.
• Individuals who shy away from handling complex data or challenging debugging tasks.
• Candidates preferring purely theoretical research without production influence.
• Seekers of low-intensity internship experiences.

Ideal Candidate Traits

• Passionate creators focused on deploying ML models into production environments.
• Systematic thinkers with a comprehensive view of the entire pipeline rather than isolated model components.
• Calm and effective troubleshooters of unexpected model behaviors.
• Self-driven individuals who prioritize real-world impact and take high initiative.