Speech / Applied ML Engineer (Intern)

We’re looking for an Applied ML Intern (Speech) who wants to work on real-world systems under real production constraints, not just experiments in notebooks.

You’ll take ownership of improving speech models used in production across Southeast Asian languages, accents, and noisy environments. This means dealing with messy data, evolving requirements, and tight constraints on latency, cost, and reliability.

You’ll work closely with engineers and founders to ship models and improvements that directly impact users. Your work won’t stay experimental. It will go live, face real-world conditions, and continuously evolve.

If you’re someone who enjoys debugging hard problems, thinking beyond metrics, and shipping meaningful ML improvements, this role will push you in the right ways.

What You Will Do

• Experiment with and improve speech/ASR models across SEA languages and accents
• Design and run experiments under real-world constraints (latency, cost, memory)
• Identify failure modes and edge cases in production speech data
• Optimize inference performance and GPU utilisation
• Develop strategies for multilingual and code-switching scenarios
• Work with engineering to deploy models into production pipelines
• Build evaluation datasets and tracking systems for model performance
• Document experiments, trade-offs, and learnings clearly

• Strong fundamentals in Python and PyTorch
• Understanding of speech/ASR basics
• Experience with model training, fine-tuning, and evaluation
• Familiarity with inference optimisation and GPU workflows
• Ability to work with messy, multilingual, real-world data
• Comfort making decisions with incomplete signals and evolving requirements

• You think in terms of shipped improvements, not just metrics
• You ask “how will this behave in production?” before trying something new
• You take ownership of speech quality and system outcomes
• You balance research depth with speed of execution
• You proactively find model failures instead of waiting for them to surface

• Experience with multilingual or low-resource speech systems
• Exposure to low-latency or on-device inference
• Experience deploying ML models into production systems

Within 4–6 weeks, you should be able to:

• Own improvements for a specific speech use case or language
• Ship at least one measurable gain in accuracy, robustness, or latency
• Identify and document key failure modes and mitigation strategies
• Contribute to evaluation, monitoring, and model diagnostics

• Hands-on experience with applied ML under real production constraints
• Direct collaboration with founders and experienced engineers
• A portfolio of shipped improvements—not just experiments
• Exposure to real-world speech challenges across languages and environments
• A strong foundation for applied ML or speech-focused engineering roles

• If you only want to work on clean datasets and offline benchmarks
• If you avoid messy data or complex debugging
• If you prefer purely research environments disconnected from production
• If you’re looking for a low-intensity internship

• Builders who enjoy shipping ML systems to production
• Engineers who think beyond models and understand full pipelines
• Calm, methodical debuggers of unpredictable system behaviour
• High-agency individuals who care about real-world impact

We’re building the speech intelligence layer for Southeast Asia—turning real-world, accented, code-switched speech into structured, usable outputs for businesses.