AI Research Intern

About this role

About the role: - Join Lexsi.ai as an AI research intern working on cutting-edge machine learning and deep learning projects - Contribute to research and development efforts spanning model architectures, interpretability, and optimization - Collaborate with a research team on problems across multiple domains of artificial intelligence What you'll work on: - Building and training transformer-based models, including architectures like BERT, GPT, LLaMA, T5, MoE, and Mamba - Implementing explainability methods such as SHAP, LIME, Integrated Gradients, LRP, Deconvolution, and Grad-CAM to make models interpretable - Exploring mechanistic interpretability through circuit analysis, activation patching, and feature visualization techniques - Developing uncertainty quantification approaches using Bayesian methods, ensemble techniques, or test-time augmentation - Optimizing neural networks through quantization and pruning to reduce model size, latency, and memory usage - Working with large language model alignment, including prompt engineering for few-shot and zero-shot scenarios, chain-of-thought reasoning, RLHF workflows, reward modeling, and human-in-the-loop fine-tuning - Experimenting with tabular foundational models such as Orion, TabPFN, and TabICL - Fine-tuning and adapting pre-trained models using full-model fine-tuning, LoRA, adapters, instruction tuning, knowledge distillation, and domain specialization What you'll need: - Strong Python skills with ability to write clean, modular, and testable code - Deep understanding of machine learning and deep learning principles with hands-on PyTorch experience - Comprehensive knowledge of transformer architectures, attention mechanisms, positional encodings, tokenization, and training objectives - Proficiency with Git workflows, CI/CD practices, packaging, documentation, and collaborative development - Excellent communication skills and experience with peer code reviews and agile teamwork What would strengthen your application: - Prior experience with any one domain: XAI methods, mechanistic interpretability, uncertainty estimation, model compression, LLM alignment, tabular foundational models, or post-training adaptation - Publications in venues like CVPR, ICLR, ICML, KDD, WWW, WACV, NeurIPS, ACL, NAACL, EMNLP, or IJCAI, or equivalent research background - Open-source contributions to AI or ML libraries and tools - Exposure to risk-sensitive applications in finance, healthcare, or related fields - Familiarity with large-scale training infrastructures and performance optimization techniques