5 months ago
London, United KingdomStaff+
Responsibilities
- Build and own end-to-end ML pipelines spanning data, training, evaluation, inference, and deployment.
- Fine-tune and adapt models using LoRA, QLoRA, SFT, DPO, and distillation.
- Architect and operate scalable inference systems while balancing latency, cost, and reliability.
- Design and maintain synthetic and real-world training data systems.
- Implement evaluation pipelines for performance, robustness, safety, and bias.
- Own production deployment, including GPU optimization, memory efficiency, latency reduction, and scaling policies.
- Collaborate with application engineering to integrate ML systems into backend, mobile, and desktop products.
- Make pragmatic trade-offs and ship iterative improvements under production constraints.
Requirements
- Strong background in deep learning and transformer-based architectures.
- Hands-on experience training, fine-tuning, or deploying large-scale machine learning models in production.
- Proficiency with at least one modern ML framework, such as PyTorch or JAX.
- Experience with distributed training and inference frameworks such as DeepSpeed, FSDP, Megatron, ZeRO, or Ray.
- Strong software engineering fundamentals and the ability to write robust, maintainable, production-grade systems.
- Experience with GPU optimization, including memory efficiency, quantization, and mixed precision.
- Ability to own ambiguous, zero-to-one ML systems end-to-end.
- Ideal experience includes LLM inference frameworks such as vLLM, TensorRT-LLM, or FasterTransformer.
- Open-source contributions to ML or systems libraries are desirable.
- Experience in scientific computing, compilers, or GPU kernels is desirable.
- Experience with RLHF pipelines, including PPO, DPO, or ORPO, is desirable.
- Experience training or deploying multimodal or diffusion models is desirable.
- Experience with large-scale data processing using Apache Arrow, Spark, or Ray is desirable.
Benefits
- Interviews are conducted via virtual meetings and/or onsite.
- The interview process typically includes 3 to 4 interviews with a prompt decision process.
- The role involves joining a small, high-talent-density, hands-on team.
Tech Stack
Categories
About Bjak
Bjak builds an online insurance marketplace for consumers in Malaysia and Southeast Asia, enabling comparison and purchase of car, health, life, and travel policies. Its business model is aggregator-based, earning commissions and fees from insurers and partners while streamlining digital applications and renewals. Founded in 2019 and headquartered in Selangor, it is privately held and is expanding from insurance comparison into broader fintech services via a planned finance super app.