Senior AI scientist

Cerence AI; India; On-site; INR 3,500,000 - 6,800,000; Full time

Job summary

Cerence AI in India seeks a scientist to join XUI‑Cerence's conversational AI platform, working with large language models and multimodal capabilities on edge and in cloud. You will design and train transformer-focused models, run experiments, and own scaling strategies across data, compute, and architecture.

You will closely partner with ML systems teams, implement stabilization techniques, and explore loss functions and RLHF variants.

Responsibilities

• Apply strong fundamentals in deep learning and representation learning.
• Design and modify transformer architectures, including attention variants, RoPE, ALiBi, GQA, MoE.
• Own optimizer and scheduler choices (AdamW, Lion, Adafactor); tune learning-rate and warmup schedulers.
• Understand and debug optimizer instability and gradient pathologies.
• Apply and validate scaling laws; navigate Chinchilla-style compute vs data tradeoffs.
• Design and experiment with loss functions including next-token prediction, contrastive objectives, RLHF, DPO, GRPO.
• Design and execute large-scale training using FSDP, ZeRO-3, tensor parallelism and pipeline parallelism; apply mixed precision and gradient checkpointing.

Job description

We are looking for a scientist to be a part of XUI‑Cerence's conversational AI platform using large language models including multimodal (both on Edge and in cloud).

What You Will Work On

• Design and train large‑scale transformer and hybrid foundation models
• Diagnose and resolve training instabilities at scale
• Navigate scaling tradeoffs across data, compute, and architecture
• Define the technical direction for next‑generation models

Core Responsibilities

Deep Learning & Transformer Foundations

• Apply strong fundamentals in deep learning and representation learning
• Design and modify transformer architectures, including:
• Attention variants
• RoPE, ALiBi
• Grouped Query Attention (GQA)
• Mixture‑of‑Experts (MoE)
• Build models from first principles, not just adapt pre‑existing codebases

Optimization Dynamics & Training Stability

• Own optimizer and scheduler choices, including:
• AdamW
• Lion
• Adafactor
• Learning‑rate and warmup schedulers
• Understand and debug optimizer instability and gradient pathologies

Scaling Laws & Compute Tradeoffs

• Apply and validate scaling laws
• Navigate Chinchilla‑style compute vs data tradeoffs

Loss Functions & Alignment

• Design and experiment with loss functions including next‑token prediction, contrastive objectives, RLHF, DPO, GRPO

Distributed Foundation Model Training

• Design and execute large‑scale training using FSDP, ZeRO‑3, tensor parallelism, pipeline parallelism
• Apply mixed precision (bf16, fp8) and gradient checkpointing
• Partner closely with ML systems teams while retaining architectural ownership

Equal Opportunity Employer

Cerence is firmly committed to Equal Employment Opportunity (EEO) and to compliance with all federal, state and local laws that prohibit employment discrimination on the basis of age, race, color, gender, gender identity, gender expression, sex, sex stereotyping, pregnancy, national origin, ancestry, religion, physical or mental disability, medical condition, marital status, citizenship status, sexual orientation, protected military or veteran status, genetic information and other protected classifications. Cerence Equal Employment Opportunity Policy Statement.