Oracle Engineer Details Full AML Agent Stack: Memory Layers, Skill Distillation, Continual Learning
AI Engineer · youtube · 2026-10-17
Nacho Martinez of Oracle presents the complete technical stack for an autonomous anti-money-laundering (AML) agent for a mid-sized bank at the AI Engineer conference. Core thesis: the model is the part you can't change; real gains come from the harness around it.
Key topics:
- Four types of agent memory: short-term, episodic, procedural, and semantic, plus handling memory conflicts
- Context engineering: summarization vs. compaction, context budget management
- Toolbox & skillbox patterns: abstracting tools and skills into reusable components; skill distillation consolidates frequent operations
- How a frozen model gets better: continual learning on fixed models and instruction hierarchy
- Engineering advice: files vs. databases, pre-built SQL vs. dynamic queries, hybrid search (semantic + keyword), keep the memory layer simple, and build small custom harnesses
More from coding & agent
- Ilya Razensteyn interviewed ~2 dozen engineers on coding agents; writeup due next week — thomasahle · 2026-10-17
- AIPOCH open-sources Open-Science: desktop app where AI agents run reproducible scientific analyses — rohanpaul_ai · 2026-10-17
- Stripe launches real-time Postgres sync for Stripe data, no webhooks needed — alexmacgregor__ · 2026-10-17
- MiMo-V2 paper: agentic groupwise grading redistributes RL advantages to curb reward hacks — Sauers_ · 2026-10-17
- Creator shares full AI music video pipeline: Claude for lyrics, Suno for song, local LTX for video — elonmusk · 2026-10-17
- One Operator Plus One Agent: A Full Prompt for Building an AI GTM Sales Team — VibeMarketer_ · 2026-10-17