AI & ML Research 3 Days

Last updated: July 12, 2026, 5:30 PM ET

AI Model Development and Challenges

Researchers are exploring methods to improve the performance and reliability of AI models. One area of focus is Retrieval Augmented Generation (RAG), a technique that augments LLMs with external knowledge. However, RAG is seen by some as a temporary solution, with future AI infrastructure potentially relying on persistent neural states and strict latency budgets rather than vector databases. Another challenge is prompt inflation in long-context LLMs, where accumulated tokens increase costs and latency; a prompt-pruning layer is proposed to address this issue a prompt-pruning layer. Hallucinations remain a problem even in frontier AI models, prompting investigations into their causes and potential solutions.

Agentic AI and Orchestration

The concept of agentic AI, where AI systems perform tasks autonomously, is being examined critically. One perspective suggests that over-reliance on external consulting mirrors the risks of delegating cognitive tasks to machines, framing agentic AI as a potential "big con". In practice, orchestrating multiple AI agents is becoming a focus, with methods demonstrated for running over 100 agents in parallel using tools like Claude Code.

Data Engineering and AI Integration

Data engineering practices are evolving to support AI workloads. Building production-ready data pipelines, such as an RSS pipeline using Python, Docker, Postgre SQL, and Kestra, requires thinking like a data engineer. For those looking to advance their skills, a practical guide covers intermediate PySpark concepts including partitions, shuffles, joins, caching, and execution plans. Deutsche Telekom is integrating AI deeply into its operations, aiming to become an AI-native telco by transforming customer service, workflows, and network operations with OpenAI.