Towards Data Science

Key takeaways The problem: point a normal chat-oriented LLM runtime at a live robot camera and three things break at once: (i) VRAM overflows from a vision stream that never ends, (ii) control-loop deadlines get missed silently, and (iii) a 60Hz camera easily outpaces a much slower reasoning step.

This series argues a specific way to build enterprise RAG, and it disagrees with a lot of standard practice along the way. RAG is not machine learning; embeddings are not magic; a chunk-size sweep optimizes the wrong thing; the answer schema matters more than the model. None of these are neutral, and each one changes what you build. This piece collects the positions the rest of the series argues from, then maps the series article by article, so you can go straight to the argument you want to check.

Most of us customize Codex through prompts. We can describe the task, give instructions, and tell Codex what kind of result we expect. This allows us to control how the agent approaches its work. But sometimes, prompting is not enough. We may want to further customize the execution by running our own logic at different stages of a Codex session. So, how can we do that? The answer is Codex hooks. In this post, we’ll explore the concept of hooks and understand where they fit into the agentic loop.

addresses the fundamental challenge of slow, sequential token generation in language model inference. DFlash speculative decoding support for CPUs was recently enabled in vLLM v0.25.0. In our testing with Qwen3.5-9B on an r8i AWS instance, powered by Intel® Xeon® 6 processors with Performance-cores, DFlash increased average token generation throughput to 3.92x that of the autoregressive baseline at concurrency 1—a 74% cost reduction per generated token.

TL;DR don’t start with a bad model. They start earlier, when instructions, evidence, memory, and tool output are flattened into ordinary strings before the prompt is built, making their original roles harder to inspect and validate. I built a small, zero-dependency Python runtime.

Here is a dataset. 432 people were released from prison and were followed for one year. They were being watched for a single event: re-arrest. Some were re-arrested in week 8, some in week 30. Most (318 of them) reached the end of the study having never been re-arrested at all. The simple question that was being asked in this study was how long until someone re-offends? We cannot just average the arrest times. Three-quarters of the people never got arrested, so they have no arrest time to average.

longer, and the advice that comes with them is simple: stop retrieving, put the whole thing in the prompt. On a case file that advice nearly works. Eleven PDFs about one insurance claim come to sixty-odd pages, which fits inside a 200,000-token window several times over. It still misses the two questions the person handling the claim has. Is the second repair invoice in the folder at all, and does the date of loss on the claim form match the one in the adjuster’s report?

1. Introduction a coding agent in action, you’ve probably noticed the default workflow: let the AI handle the routine bugs, but step in when the problem gets complicated. One-line fixes go to the agent; deep internals, numerical edge cases, and cross-file invariants stay with the senior engineer. That made me wonder: what kinds of bugs do coding agents actually struggle with?

of this series, I built a stateful LangGraph agent that handles a 15-minute booking process and wrapped it up with a Streamlit UI to improve user experience. The agent handles the entire booking process like a real customer service representative. It’s a LangGraph-based agent that orchestrates the following operations: Responds to customer queries and understands their needs. Calculates the price for the service and informs the customer. Handles the customer’s acceptance or rejection.

for taking a RAG system past one document is to build an index: one row per document, one column per field you may want to filter on, then a filter in front of the search. That advice assumes the documents have fields in common. Plenty of folders do not. Take a research folder: a 492-page catalogue of security controls, a zero-trust architecture spec, an AI risk framework, thirty-one machine-learning papers, and seven commodity market reports. Ask what column they share.