ML Intern – Fine-Tuning & Model Adaptation
Internship · 10–12 Weeks · 2 Openings · San Francisco, CA
About CompFly AI
CompFly is building the security layer for the autonomous AI era; real-time defense for AI agents that act, decide, and operate on behalf of humans inside enterprise environments. As autonomous agents become the new compute primitive, CompFly ensures enterprises can trust what those agents do. We are a small team that moves fast and builds things that matter.
The Role
We are hiring two ML interns to join our model adaptation team. Each intern will own a distinct workstream within our fine-tuning and post-training pipeline, working hands-on to specialize foundation models for agentic security applications. You will go well beyond prompting; you will be in the training loop, running experiments, analyzing results, and shipping model improvements that affect production. These roles are ideal for ML students with strong fundamentals who want deep, practical experience with the model training stack in an applied security context.
What You’ll Work On
- Fine-tune and instruction-tune foundation models using LoRA, QLoRA, and other PEFT methods
- Design ablation studies across data composition, learning rate schedules, and regularization strategies
- Build evaluation harnesses and synthetic data pipelines to benchmark model performance on limited labeled datasets
What We’re Looking For
- Currently enrolled in a BS, MS, or PhD in Machine Learning, Computer Science, or a related field.
- Solid understanding of transformer architectures, attention mechanisms, and the supervised fine-tuning process.
- Proficiency in Python; hands-on experience with PyTorch and the HuggingFace ecosystem (transformers, datasets, PEFT, accelerate).
- Experience tracking and analyzing ML experiments (W&B, MLflow, or equivalent).
- Strong experimental mindset, you formulate clear hypotheses, control variables, and communicate findings precisely.
- Ability to work independently on a defined research workstream with regular team syncs.
Nice to Have
- Experience with RLHF, DPO, KTO, or other preference optimization and alignment techniques.
- Prior internship or research project involving LLM fine-tuning, post-training, or model evaluation.
- Interest in adversarial ML, model robustness, or AI security applications.
What You’ll Gain
Both interns will own a meaningful workstream, present results to senior researchers, and contribute work that ships into production. Exceptional interns will be considered for full-time roles. You will leave with deep hands-on experience fine-tuning frontier models in a real applied security setting, a rare combination.
Compensation: $40-50 per hour
How to Apply:
E-mail resume at careers@compfly.ai