Aakash Bist

Aakash Bist

Staff Software Engineer, Office of the CTO

About

Aakash is a Staff Software Engineer at Everpure focused on building AI systems and agentic workflows. His work spans storage-accelerated LLM inference, scalable agent architectures and explainable human-in-the-loop AI systems.

LLM agents often fail at enterprise scale because we treat them as data processors rather than orchestrators. Facing petabytes of production telemetry and billion-row tables, the standard "Step-and-Loop" architecture collapses under latency, security risks, and context window bloat. This talk explores a shift from reactive agents to declarative planners. We detail how we built a high-scale system that separates reasoning from execution. By using jq as a safe, sandboxed transformation layer, our agents generate complete, auditable execution plans upfront keeping massive datasets out of the context window while maintaining strict security.

Read More →