Topics

Systems where a language model plans, uses tools, takes actions over several steps or works with other agents to complete a task — including how well they work, how they fail, and how companies deploy them.

Giving language models access to outside documents or databases at answer time — retrieval, ranking, chunking, grounding, citing sources, and how well these systems answer from company knowledge.

Language models that write, fix, review, test or explain software — coding assistants, code agents, program repair, and how they are measured on real software work.

Vision for manufacturing and industrial settings: defect detection, visual inspection, industrial anomaly detection, quality control on production lines. Not autonomous driving, general robotics or remote sensing.

How language models are tested — new benchmarks and datasets, problems with existing benchmarks (contamination, saturation), LLM-as-judge, and evaluation methods companies can trust.

Making models cheaper and faster to run — small language models, distillation, quantization, pruning, efficient inference and serving, and on-device models.

Models that work across text, images, video, audio or documents together — vision-language models, document understanding, and multimodal reasoning.

Whether AI systems behave safely and reliably once in use — hallucination, jailbreaks and prompt injection, robustness, monitoring, guardrails, red-teaming, and alignment methods that matter for real deployments.