Founding ML Engineer
Bolna AIVoice AI company
Bangalore, IndiaSenior
Y Combinator
General Catalyst
Upekkha
Blume Ventures
Orange Collective
Pioneer Fund
Data & AI
About the role
TL;DR
Build, evaluate, and deploy machine learning models for voice AI.
- •At Bolna, we're building tools that change the way teams leverage Voice AI.
- •We're looking for a Founding Machine Learning Engineer to own the end-to-end lifecycle of building, evaluating, deploying, and improving models that power millions of production conversations.
- •Key Responsibilities Build the data engine
- •Design pipelines to source and clean conversational voice data across Indian languages, accents, and telephony conditions.
- •Fine-tune models that ship
- •Fine tune and train models to improve accuracy, speed, and reliability across different use-cases.
- •Define what "good" means
- •Build evaluation datasets and benchmarks for transcription accuracy, voice naturalness, interruption handling, latency, and end-to-end conversation quality.
- •Set up human-in-the-loop pipelines to capture subjective quality at scale.
- •Ship to production
- •Work with the engineering team to deploy models into a latency-sensitive, high-volume system.
- •Monitor performance in the wild, debug regressions, and iterate fast.
- •Requirements 3+ years of hands-on ML experience with deep practical real-world experience in training models.
- •Strong Python and PyTorch fundamentals with exposure in distributed training, and modern fine-tuning techniques (LoRA, QLoRA, DPO, RLHF, etc.).
- •Training data as a first-class problem.
- •Experience designing data pipelines from collection, cleaning, labeling, deduplication, augmentation and treating data quality as a core engineering discipline.
- •Rigorous about evaluation.
- •You know that "looks good in a demo" is not a benchmark.
- •You build the evals before you trust the model.
- •Speech model experience is a plus with real-time / streaming inference experience where you would have contributed to latency optimization, quantization, and distillation for production deployment.
- •Bias toward shipping.
- •You'd rather have a model running in production this week than a perfect one in a notebook next quarter.
Required skills
PythonPyTorchPandasNumPyAirflowAWSKubernetesCI/CDDocker
Domain expertise
ai
Benefits & perks
Competitive package, ESOP, Health Insurance
Tech stack
PythonPyTorchPandasNumPyAirflowAWSKubernetesCI/CDDocker