Deloitte

A filtered query with k=20 returns only 4 hits even though 50,000 vectors pass the filter. Pick the fix.

Design an audit to verify a vendor's claim that one expert 'handles math.'

Predict per expert capacity for 4096 tokens with 8 experts, top-2 routing, capacity factor 1.25.

Pick which workloads move onto your own GPUs in a hybrid hosted plus self-host setup

Design a code review bot for a 500 engineer org that posts within minutes

Explain when 'citations required' belongs in the output rail

Walk through an online sampled eval setup for safety in production

Reasoning over a 200 page document or a long video: what breaks, and what are the escape hatches?

Where the OpenAI Agents SDK puts input and output guardrails relative to the agent loop

Why is tokenizer training a first order pretraining choice?