ML Researcher / Engineer - Voice and Speech to Speech Models

Location: Bengaluru (Work from Office - Domlur)

Team: AI & Machine Learning

Experience: 2 –7 years
What You'll do:

\n
• Fine-tune and deploy LLMs, TTS, STT, and voice models for use in real-time conversations with millions of users.
\n

\n
• Convert unstructured, messy real-world audio/text data into clean, high-quality datasets for training and evaluation.
\n
• Build inference pipelines optimized for low-latency, high-accuracy voice agents and multimodal interfaces.
\n
• Work closely with infra and product teams to ship production-grade GenAI models with observability, fallback, and monitoring.
\n
• Experiment with GANs, diffusion models, audio generation , and multimodal fusion to power next-gen AI agents.
\n
• Own the full model lifecycle — from research and training to deployment, testing, and iteration.
\n

What we're Looking for:

\n
• 2-7 years of hands-on experience in AI / ML roles, ideally in startups or product-driven teams.
\n
• Strong grasp of LLM fine-tuning , instruction tuning, or pretraining techniques.
\n
• Familiarity with TTS/STT systems , Whisper, Tacotron, VITS, or other open source models .
\n
• Experience with multimodal architectures , generative audio, GANs, or diffusion-based models.
\n
• Ability to work with real-world messy data , design training pipelines, and debug model failure modes.
\n
• Fluency in frameworks like PyTorch, HuggingFace, TensorFlow , and ecosystem tools (ONNX, Triton, LangChain, etc.).
\n
• Passion for building high-impact AI features that ship to real customers.
  
\n

Requirements

Why Join Us:

\n
• Work at the cutting edge of LLMs, voice AI, and generative models — and ship real products, not just prototypes.
\n

\n
• Directly impact millions of users by powering AI agents that help with hiring, learning, and career growth.
\n
• Collaborate with a world-class team of AI engineers, researchers, and product minds who move fast and ship boldly.
\n
• Freedom to explore: Own experiments, propose architecture, or contribute to foundational model training.
\n
• Startup speed, enterprise scale — best of both worlds. Rapid iteration and direct customer feedback.
\n
• Multilingual India - first problems that push the boundaries of speech, reasoning, and personalization.
\n