Architecture AI Updates: November 17, 2026
Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled
1. Uber’s Legal Redlining Agent Combines RAG Over Lawyer Decisions With Deterministic Rules and Cuts Review Time by More Than 20%
Uber described a Legal Redlining Agent that sits in Microsoft Word, interprets contract edits proposed by outside parties, compares them with legal policy and suggests whether to accept, reject or modify. According to InfoQ, the first RAG design over legal playbooks struggled with retrieval accuracy and negotiation tone, so Uber added a feedback loop that stores lawyers’ decisions, reasoning and final text. At runtime, similarity search and metadata filtering narrow thousands of past interactions to about 20 candidates, a second LLM stage scores relevance, and exponential decay with a 365-day half-life favors recent decisions. Uber reports average review time down by more than 20% and 91% accuracy, with lawyers still approving every recommendation. Source
2. Instacart Splits Search Query Understanding Into Categories, Fallbacks and Extracted Attributes Using LLMs
ByteByteGo summarized how Instacart rebuilt query understanding, the first of three search stages before retrieval and ranking, with LLMs. The article covers how a general LLM is adapted to Instacart’s catalog, how a query is mapped to product categories, how the system ensures something is always suggested, how details such as dietary attributes are extracted from phrases like “bread no gluten”, and how the process is made faster. It notes that information produced at the query understanding stage guides both retrieval and ranking. Source