1 day ago
Taipei, TaiwanSenior
Responsibilities
- Evaluate Qwen, Llama, Gemma, and other open-weight foundation models for AI-phone use cases.
- Identify when existing models are sufficient and when model adaptation or fine-tuning is required.
- Experiment with SFT, LoRA/QLoRA, and other lightweight fine-tuning techniques.
- Improve model performance for intent recognition, instruction following, tool calling, structured outputs, and multi-step task execution.
- Generate synthetic data and construct datasets for phone-specific workflows.
- Build evaluation datasets and benchmarks for model quality, reliability, and task completion.
- Compare local and cloud inference approaches and collaborate with the Agent Runtime team on real agent workflows.
Requirements
- Strong background in Machine Learning, Deep Learning, or NLP.
- Practical experience working with LLMs and open-weight models.
- Experience with model fine-tuning, including SFT, LoRA/QLoRA, and preferably RL or preference optimization.
- Experience with PyTorch and common ML tooling.
- Understanding of LLM evaluation, benchmarking, and dataset preparation.
- Ability to run experiments, analyze model failures, and translate findings into measurable improvements.
- Experience with agentic systems, function calling, tool-use training, synthetic data generation, preference datasets, model distillation, smaller language models, or production model deployment is preferred.
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.