Key Responsibilities:

- Design and build scalable data pipelines for collecting, cleaning, labeling, and processing conversational voice data.

- Fine-tune and train speech and language models for improved accuracy, latency, and production performance.

- Develop evaluation datasets, benchmarks, and automated testing pipelines for speech quality and model performance.

- Build human-in-the-loop evaluation workflows to continuously improve model quality.

- Collaborate with engineering teams to deploy models into high-volume, latency-sensitive production environments.

- Monitor production model performance, investigate regressions, and continuously optimize model behavior.

- Improve speech recognition quality across multiple Indian languages, accents, and telephony conditions.

- Optimize inference performance through quantization, distillation, and latency optimization techniques.

Required Skills:

- 3+ years of hands-on Machine Learning experience with production model development.

- Strong programming skills in Python and PyTorch.

- Experience with distributed training and modern fine-tuning techniques such as LoRA, QLoRA, DPO, and RLHF.

- Strong understanding of ML data pipelines including data collection, preprocessing, labeling, deduplication, augmentation, and quality validation.

- Experience designing robust model evaluation frameworks and benchmarking systems.

- Strong understanding of production ML systems and model deployment.

- Excellent analytical and problem-solving skills.

Preferred Skills:

- Experience with Speech AI, ASR, TTS, or Voice AI systems.

- Knowledge of real-time or streaming inference architectures.

- Experience with model optimization techniques including quantization, pruning, and distillation.

- Familiarity with GPU training, distributed computing, and cloud ML infrastructure.

- Experience deploying ML models at scale.