Machine Learning Engineer III

On-site, Bangalore

LatentForce

About the Role

We are building specialized LLMs that understand and reason over massive enterprise codebases. This is real model training — RL loops, PEFT, verifiable rewards, long-context modeling — not API integration. You’ll own end-to-end experimentation and work directly with founders.

Responsibilities

• Train LLMs using RL (PPO/GRPO/RLHF/RLVR) and PEFT (LoRA, QLoRA, DoRA, IA3).
• Build custom training loops with PyTorch, HuggingFace, TRL, Unsloth .
• Design reward functions and verifiers for code-understanding tasks.
• Run full-stack ML experiments: data → training → eval → iteration.
• Develop scalable training infra (FSDP/DeepSpeed, distributed training).
• Build evaluation suites for reasoning and code comprehension.

Minimum Qualifications

• 3+ years of real deep learning experience (actual model training).
• Strong fundamentals: linear algebra, probability, optimization, statistics .
• Proven experience training transformers or large DNNs from scratch or checkpoints.
• Proficiency in PyTorch , HuggingFace , TRL , Unsloth .
• Experience implementing RL algorithms or custom training pipelines.
• Research exposure (publications/preprints) or strong open-source work.
• Ability to debug training issues (NaNs, KL drift, reward hacking, etc.).
• Startup mindset; comfortable with fast, on-site, high-performance execution.

Nice to Have

• DeepSpeed/FSDP, model parallelism, vLLM.
• Program analysis / AST tooling.
• Long-context modeling experience.

Why Join Us

• Build specialized LLMs at a well-funded early-stage company.
• Direct work with founders; high ownership and technical depth.
• High-impact role shaping core training architecture.

Apply here: