Advanced AI Interview Lab
Work through RAG failures, enterprise retrieval, KV-cache capacity, large-scale serving, GPT-style training, post-training and reasoning budgets. Eight worked lessons with interview answers and self-checks.
Advanced AI Interview Lab
Work through RAG failures, enterprise retrieval, KV-cache capacity, large-scale serving, GPT-style training, post-training and reasoning budgets. Eight worked lessons with interview answers and self-checks.
RAG, LLM Systems & GPT Design
Reference designs, explicit assumptions, capacity calculations and failure analysis. Sources reviewed 16 October 2026.
Advanced RAG: retrieval, reranking and grounded answers
A relevant-looking passage is not necessarily enough to answer a question. Diagnose failures at ingestion, retrieval, ranking and generation before changing the model.
Enterprise RAG: permissions, freshness and index migrations
Design retrieval that respects changing permissions and document versions. A good answer delivered to the wrong person is a system failure.
LLM serving: KV cache, GQA and memory budgets
Estimate what fits on an accelerator before promising a concurrency target. Separate model weights, KV cache, activations and runtime reserve.
LLM serving: prefill, decode and speculative sampling
Optimize the stage that limits the user experience. An improvement in aggregate tokens per second can still make an interactive service feel slower.
Design a GPT-style decoder and training pipeline
Start from tokenization and a causal objective, then work through training memory, distributed execution and evaluation. This is a reference decoder design, not a claim about private GPT internals.
Post-training, preference learning and release evaluation
Choose the intervention that matches the failure. More supervised data, a preference objective and a better evaluation set address different problems.
Frontier systems: MoE, reasoning budgets and tool execution
Distinguish sparse computation, longer inference and tool access. Each changes a different resource or capability and introduces a different failure mode.
Design a large-scale GPT-style chat service
Turn model inference into a reliable product: conversations, streams, admission control, regional failure handling and a release process that protects users.