Software Intern (LLM Fine-Tuning)

Role overview As a fall/winter software intern, help adapt open-weight language models for security and identity tasks. The work spans the full experiment cycle: preparing data, fine-tuning models on GPUs, and building evaluations to determine whether each change improves results.

Responsibilities - Gather, clean, and prepare datasets for training and evaluation. - Fine-tune open-weight language models using supervised fine-tuning and parameter-efficient approaches such as LoRA or QLoRA. - Run and monitor GPU training jobs while working within compute limits. - Build evaluation sets and benchmarks that reflect relevant tasks. - Compare models and training runs, then communicate findings, including what did not work.

Requirements - Working knowledge of PyTorch and Hugging Face tools such as Transformers, Datasets, and PEFT. - Hands-on experience with open-weight language models and fine-tuning methods including SFT, LoRA, or QLoRA. - Experience preparing training and evaluation data and assessing model performance. - Understanding of GPU constraints, including memory, batch sizing, and mixed precision.

Nice to have - Experience with domain-adaptive or continued pretraining, synthetic data, CUDA, Docker, or vLLM. - Familiarity with experiment tracking, evaluation frameworks, or reinforcement learning for language models.

Benefits and work setup - Fall/winter internship with remote or hybrid work in the Greater Toronto Area. - Hands-on access to real models and GPU workloads, with close collaboration with the founding team. - The work offers exposure to applying AI in security and cryptographic identity.