Pocket FM is where audio storytelling comes to life, powered by cutting-edge AI. With a global community of over 280 million listeners, we are building the world's largest platform for immersive audio series.

Our Generative AI team is pushing the boundaries of large language models, reinforcement learning, and multimodal AI to transform how stories are created, adapted, and experienced globally.

We're looking for Research Engineers who don't just experiment with AI but train models, develop new capabilities, and turn research breakthroughs into real-world products.

The Role: What You'll Build and Own

As a Research Engineer, you'll work on challenging problems in foundation model training, reinforcement learning, and post-training, developing AI systems that power next-generation storytelling experiences.

- Train & Adapt Foundation Models: Design and execute large-scale model training and fine-tuning pipelines for LLMs such as Llama, Qwen, and Mistral. Work across data preparation, supervised fine-tuning (SFT), optimization, distributed training, and model evaluation
- Advance Reinforcement Learning & Post-Training: Develop and implement reinforcement learning and preference optimization techniques, including RLHF, PPO, GRPO, DPO, and reward modeling, to improve model reasoning, instruction following, controllability, and generation quality
- Build Novel LLM Capabilities: Experiment with model architectures, training strategies, and post-training methodologies to improve long-context understanding, narrative consistency, creative generation, and multilingual capabilities
- Develop Multimodal AI Systems: Research and build models that work across text, speech, audio, and vision, enabling immersive storytelling, content understanding, and generation
- Design Agentic AI Systems:



Develop intelligent workflows involving planning, reasoning, tool use, memory, and multi-agent collaboration. Explore reinforcement learning approaches to improve agent performance on complex, long-horizon tasks
- Build Evaluation & Training Infrastructure: Develop scalable experimentation pipelines, automated benchmarks, reward functions, and evaluation frameworks. Optimize training efficiency using distributed computing frameworks
- Translate Research into Production: Read, reproduce, and extend recent research, collaborating with scientists and engineers to deploy reliable AI capabilities at scale.

The Ideal Candidate: Who You Are

We're looking for technically strong researchers and engineers who combine deep ML fundamentals with hands-on model development experience.

- Strong ML Foundation: Bachelor's or Master's degree in Computer Science, AI, Machine Learning, or a related field. A PhD is not required
- Hands-On Model Training Experience: Ideally 3 years of experience developing, training, fine-tuning, or deploying machine learning models, with demonstrated experience working directly with LLM training pipelines
- Reinforcement Learning & Post-Training Expertise: Practical experience with reinforcement learning, reward modeling, preference optimization, SFT, or RLHF. You understand the underlying algorithms and can implement and experiment with training techniques beyond simply using existing APIs




- Deep Understanding of LLMs: Solid knowledge of transformers, attention mechanisms, tokenization, optimization, pre-training, post-training, and model evaluation
- Research & Experimentation Mindset: Ability to understand research papers, reproduce results, design experiments, and improve existing approaches through systematic evaluation
- Production Engineering Experience: Experience building scalable ML systems and taking models from experimentation to production.

Your Technical Toolkit

- Required: Strong proficiency in Python, PyTorch, and the Hugging Face ecosystem (Transformers, TRL, Accelerate, PEFT)
- Model Training: Hands-on experience with SFT, distributed training, checkpointing, optimization, and model evaluation
- Reinforcement Learning & Post-Training: Experience with RLHF, PPO, GRPO, DPO, reward modeling, or related reinforcement learning and alignment techniques
- Distributed Training: Familiarity with FSDP, DeepSpeed, Megatron-LM, Ray, or similar frameworks
- Bonus Points: Research publications at NeurIPS, ICML, ICLR, ACL, or EMNLP; experience with multimodal models, agentic systems, multilingual LLMs, or novel evaluation frameworks.

Why Join Us?

- Work on Frontier AI Problems: Solve challenging problems in foundation model training, reinforcement learning, post-training, and generative AI
- Research That Ships: See your work move beyond experiments and directly influence products used by millions of listeners
- Ownership & Autonomy: Take ownership of meaningful research and engineering problems, from experimentation to deployment
- Top-Tier Rewards: Competitive compensation, meaningful employee stock options (ESOPs), and opportunities to grow within a rapidly evolving AI organization.

📌 Research Engineer – Applied Generative AI (LLMs & Multimodal Systems) (Bengaluru)
🏢 Pocket FM
📍 Bengaluru